Light-sensing device

Description of equivalent WO 2004027879 A2

LIGHT-SENSING DEVICE

0002Background of the invention

0003The present invention relates to methods of optical to electronic conversion, in particular to photo- detection with devices suitable for integration with Complementary Metal Oxide Semiconductor (CMOS) technology.

0004The devices and fabrication technologies used for light-sensing are highly dependent on the frequency or wavelength of the light to be sensed. Usually, for imaging purposes, the spectrum is divided into:

0005Ultra-Violet (UN), wavelengths shorter than 0.4μm Visible, wavelengths in the range of 0.4μm to 0.7 μm Short-Wave Infra-Red (S WIR), wavelengths in the range of 1 μm to 3 μm

0006Mid- Wave I fra-Red (MWIR), wavelengths in the range of 3μm to 5μm Long- Wave Infra-Red (LWIR), wavelengths in the range of 8μm to 12μm Nery Long-Wave Infra-Red (VLWIR), wavelengths longer than 20μm

0007For wavelengths in the visible range, a wide range of devices and materials can be used, with the most common types being silicon pn-junction photodiodes, either in CCDs or CMOS image sensors. For UN detection, it is usually desirable to have "solar-blind" detection, i.e., absorption of UN without absorption of the visible wavelengths. This can be achieved with pn-junction or heterojunction photodiodes made with materials whose bandgap energy is larger than the energy of the photons with wavelengths in the visible range. Examples of such materials are Silicon-Carbide (SiC), GaΝ, AlGaΝ, A1Ν, ZnO, etc. It is also possible to have solar-blind detection with thinned CCDs. These devices make use of the fact that the coefficient of absorption of most materials, including silicon, increases for shorter wavelengths. Therefore, solar-blind UN detection is achieved by reducing the thickness of the light sensing material, so that the signal generated by visible light is much weaker that the signal generated by UN light. Yet another possibility for solar-blind UN detection is to place a visible-blocking filter in front of the light-sensing elements.

0008The sensing of the different IR bands of wavelengths can be approached with very different devices, operating under different physical mechanisms, typically with very different materials systems. The SWIR band includes the 1.310μm and the 1.550μm ranges of wavelengths that are used for fiber-optics communications. These wavelengths can be absorbed by pn-junction or heterojunction photodiodes made with Germanium (Ge) or Indium-Phosphide (InP). It has been shown that SiGe and/or SiGeC random alloys and/or superlattices can also detect these signals.

0009The MWIR band can be approached by devices operating on band-to-band transitions, such as pn- junction and heterojunction photodiodes, provided that the materials used have a small enough bandgap, for example HgCdTe. Other devices, such as Quantum Well Infrared Photodetectors (QWIPs) can also be used. QWIP devices operate with intersubband transitions, i.e., transitions between discrete energy levels either in the conduction-band or in the valence-band of quantum wells. Some common materials employed are GaAs-AlGaAs, InGaAs-InAlAs, InGaAs-InP, GaAs- GalnP, GaAs-AlInP, etc. Another physical mechanism employed for the detection of these wavelengths is the Heterojunction Internal Photoemission (HIP). This mechanism can be implemented with semiconductor-semiconductor heterojunctions or with metal-semiconductor heterojunctions (also known as Schottky-diodes). The barrier height at the heterojunction determines the longest wavelength possible to detect. QWIPs and HIP devices are unipolar and can be designed to operate with either electrons or holes. The perfonnance of all these different types of devices can be significantly increased by lowering the temperature of operation. Typical temperatures for the HIP devices are around 77K (liquid nitrogen cooling).

0010The LWIR band can be sensed with essentially the same devices used for the MWIR, but with the appropriate fine-tuning of the device-parameters. For devices based on band-to-band transitions this requires a decrease in band gap energy, for QWIP devices this requires the energy difference between the quantized levels to be decreased, that is, the width of the quantum well to be increased, and for HIP devices the heterojunction or Schottky "barrier height" needs to be decreased. All these result increased noise and decreased perfonnance of the light-detection, which can be counter-acted by further decreasing the operating temperature.

0011The NLvVTR band can be covered by some of the devices mentioned for LWIR, including HIP devices, but all need to operate at very low temperatures of around 4K (liquid helium cooling). In addition, there is another physical mechanism that is used for these wavelengths: excitation of doping impurities. The energy levels of doping elements are usually very close to either the conduction band or the valence band of the host semiconductor (Si, Ge, GaAs, etc.). At room temperature the doping impurities can be taken to be fully ionized. The energy difference between in the "band of impurities" and the host semiconductor corresponds to the energy of photons in this band. At very low temperature, the impurities remain neutral and the host semiconductor remains "intrinsic". Therefore, photons can excite the impurities and inject carries into the energy bands of the host semiconductor, thereby producing a detectable signal.

0012Multispectral and hyperspectral imaging, covering at least some of the spectral bands mentioned above, offer many interesting possibilities for the retrieval, processing and subsequent display, of data that the human eyes cannot perceive, and that can have a wide range of applications. It has been very challenging to fabricate monolithically integrated sensors for the different wavelength bands (multi-spectral sensors), having pixel density and cost of fabrication comparable with the state of the art commercial CCDs and CMOS image sensors. The main reason for this situation is the lack of suitable device/materials and process architectures capable of fabricating sensors covering most of the spectral ranges mentioned above.

0013Image sensing in different ranges of wavelengths requires different types of photo-detectors. The differences between photo-detectors range from materials system, device physics, device architecture, mode of operation, etc. The reason behind this diversity in photo-detection technologies is due to, firstly, all the possible different approaches to tackle the physics involved in the absorption of photons of a particular wavelength, and secondly, what are the best technologies for each range of wavelengths of interest.

0014Conventional photo-detectors covering the wavelengths in the visible range, such as Charge Coupled Devices (CCDs), CMOS photodiodes and CMOS photo-gates are made on silicon substrates and involve what are called "band to band transitions" in which electron-hole pairs are generated. For this type of photo-detection a photon is absorbed by scattering one electron from the valence band to the conduction band, thus creating a "hole" in the valence band.

0015There is strong interest in "solar-blind" or "visible-blind" UN photo-detection for which photo- detectors made with wide band-gap materials such as SiC (Silicon Carbide) or GaΝ (Gallium Nitride) have been demonstrated. Such materials have band-gaps larger than the energy of the photons in the visible range, and therefore band-to-band transitions cannot take place. Photons in the UN range of the spectrum have energy larger than the band-gap of these materials and can therefore generate electron-hole pairs through band-to-band transitions.

0016hi the Infra-Red (IR) portion of the spectrum, the situation is little bit more complicated, as there are several distinct regions of the spectrum. Different technologies are under development to improve the performance for each of these wavelength intervals. For example, Silicon devices such as

0017CCDs, pn-junction diodes or p-i-n (PDSf) junction diodes, are capable of absorption in the SWIR band (0.7μm < λ < lμm) and germanium devices can absorb wavelengths up to 1.6μm. For the MWIR range (3μm < λ < 5μm), Heterojunction Internal Photoemission (HIP) detectors making use of Schottky-junctions between p+-Silicon and PtSi (Platinum Suicide) seem to offer good performance. On the other hand the best performing devices and materials for LWIR (8μm < λ < 12μm) seem to be HgCdTe (Mercury Cadmium Telluride - MCT) photodiodes. For LWIR SiGe/Si HIP devices have also been fabricated but their performance seems considerably lower than that of MCT-based devices. As a result, sensors made with MCT can operate without cooling while the best SiGe/Si HIP devices require cooling to 77K.

0018The fact that each range of wavelengtli requires a different type of photo-detector and/or a different materials system, has made impossible the fabrication of ID and or 2D arrays (or focal planes) of "color" pixels capable of capturing multiple-wavelengths in the visible and invisible parts of the spectrum, such as UN, Visible, SWIR, MWIR, LWIR.

0019Conventional solutions include the sensing with separate cameras each with its own set of lenses, focal planes, along with all the required circuitry to produce still images and/or video signals. Hybrid schemes have been developed, in which there is a single set of lenses, but still require the splitting of the incoming light through prisms and wavelength filters onto different focal planes with the type of photo-detectors suitable for each desired range of wavelengths.

0020Having a single set of lenses and a single focal plane capable of image sensing in all the desired wavelengths would present an extraordinary advancement towards the goal of "sensor fusion". Therefore, it is of extreme relevance to develop a process flow for the fabrication of photo-detector devices suitable for monolithic integration with CMOS circuitry for ID and/or 2D arrays of Passive- Pixel Sensors (PPS), and/or ID and/or 2D arrays of Active-Pixel Sensors (APS), in which each pixel has multi- wavelength (visible and invisible) sensing capability. The present invention shows how to achieve this goal.

0021Summary of the Invention

0022It is an object of the present invention to provide a method of fabricating light-sensing devices including photodiodes monolithically integrated with CMOS devices on a common semiconductor substrate.

0023Anotlier object of the invention is to provide a method of fabricating light-sensing devices including Two or more types of photodiode devices integrated with CMOS devices on a common semiconductor substrate and allowing the operational features of the devices to be tailored to suit various wavelength sensing requirements.

0024Yet another object of the invention is to provide a method of fabricating photo-detector devices including PIN and HIP photodiode devices integrated with CMOS devices on a common semiconductor substrate and allowing the photodiodes to be operated under Avalanche and Geiger modes.

0025A further object of this invention is to provide a method of fabricating a photo-detector device, in which the same device can be operated as a PIN photodiode when a certain bias condition is applied, or can be operated as a HIP photodiode when another particular bias condition is applied.

0026A further object of this invention is to provide a method of fabricating HIT cooling devices, fabricated directly on photo-detector devices including PIN and HIP photodiode devices, all integrated with CMOS devices on a common semiconductor substrate.

0027According to the present invention, two or more types of photodiode devices are simultaneously epitaxially grown on side-by-side active areas implanted in a common semiconductor substrate that also includes CMOS devices, the active areas being electrically isolated from one another and from the adjacent CMOS device. The photodiode devices are grown on surfaces having defined doping polarities. The epitaxially grown layers for the photodiode devices may be either undoped or in-situ doped with profiles suitable for their respective operation.

0028With appropriate choice of substrate materials, device layers and heterojunction engineering and process architecture, it is possible to fabricate silicon-based and germanium-based multi-spectral sensors that can deliver pixel density and cost of fabrication comparable to the state of the art CCDs and CMOS image sensors. The present invention can be implemented with epitaxially deposited films on the following substrates: Silicon Bulk, Thick-Film and Thin-Film Silicon-On-Insulator (S OI), Germanium Bulk, Thick-Film and Thin-Film Germanium-On-Insulator (GeOI).

0029Photo-detector devices grown according to the invention are suitable for the realization of ID and 2D arrays of Passive-Pixel Sensors (PPS), and also for the realization of ID and 2D arrays of Active-Pixel Sensors (APS), in which each pixel has multi-wavelength (visible and invisible) sensing capability. Also, the invention permits to realize photo-sensing devices for front-side illumination as well as photo-sensing devices suitable for back-side illumination. Brief description of the drawings

0030Figure 1 shows a combination of PIN and HIP devices, side-by-side, monolithically integrated with CMOS devices, in which the PIN device is made on n+ active area. This configuration is meant for Front-Side Illumination.

0031Figure 2 shows a combination of PIN and HIP devices, side-by-side, monolithically integrated with CMOS devices, in which the PIN device is made on p+ active area. This configuration is meant for Front-Side Ilhnnination.

0032Figure 3 shows a combination of PIN and HIP devices, side-by-side, monolithically integrated with CMOS devices, with in-situ p-type doping of the initial epitaxial layers, and Schottky junctions as top electrodes. This configuration is meant for Back-Side Illumination.

0033Figure 4 shows a combination of PIN and HIP devices, side-by-side, monolithically integrated with CMOS devices, with in-situ n-type doping of the initial epitaxial layers, and Schottky junctions as top electrodes. This configuration is meant for Back-Side Illumination. Figures 5 to 8 show four different geometries with which it is possible to implement the same device (PIN or HIP).

0034Figures 9 and 10 show schematic representations of the ideal energy-band diagram of avalanche layers that under a certain bias provide electron-only avalanche multiplication, and under the reverse polarity bias provide hole-only avalanche multiplication, as long as both types of carriers are injected from the same direction into the avalanche layers.

0035Figures 11 and 12 show schematic representations of avalanche layers and associated energy-band diagrams, implemented with SiGeC random alloys, or (Siι_-yCy)_m-(Geι_-xCχ)_n short-period strained-layer superlattices on Si substrates.

0036Figures 13 and 14 show schematics of the ideal heterojunction and doping profiles, along with a schematics of the corresponding energy-band diagrams, of Avalanche PIN devices for Si substrates using SiGeC random alloys.

0037Figures 15 and 16 show schematics of the ideal heterojunction and doping profiles, along with a schematics of the corresponding energy-band diagrams, of Avalanche PIN devices for Si substrates using (Siι_-yCy)_m-(Geι__xC_x)_n short-period strained-layer superlattices. Figure 17 shows the ideal energy-band diagram, heterojunction and doping profiles of p- type Avalanche HIP devices for Si substrates using Siι_-xι-yiGe_xιC_yi/Siι_-X2-y₂Ge_J!2Cy2 random alloys.

0038Figure 18 shows the ideal energy-band diagram, heterojunction and doping profiles of n- type Avalanche HIP devices for Si substrates using Si Siι_-yC_y random alloys.

0039Figure 19 shows the ideal energy -band diagram, heterojunction and doping profiles of p- type Avalanche HIP devices for Si substrates using (Siι_-yCy)_m-(Geι_-xC_x)_n short-period strained-layer superlattices. Figure 20 shows the ideal energy-band diagram, heterojunction and doping profiles of n- type Avalanche HIP devices for Si substrates using (Siι_-yC_y)_m-(Geι._xC_x)n short-period strained-layer superlattices.

0040Figures 21 and 22 show the energy-band diagrams for combined PIN & HIP (p-i-p) devices side-by-side, as shown is Figure 1, in which the epitaxial layer stack is designed for Front-Side Illumination, with p-type absoφtion layers.

0041Figures 23 and 24 show the energy-band diagrams for combined PIN & HIP (n-i-n) devices side-by-side, as shown in Figure 2, in which the epitaxial layer stack is designed for Front-Side Illumination, with p-type absoφtion layers. Figures 25 and 26 show the energy-band diagrams for combined PIN & HIP (p-i-metal)

0042Schottky devices, as shown is Figure 3, in which the epitaxial layer stack is designed for Back-Side Illumination, with p-type absoφtion layers.

0043Figures 27 and 28 show the energy-band diagrams for combined PIN & HIP (n-i-metal) Schottky devices, as shown in Figure 4, in which the epitaxial layer stack is designed for Back-Side Illumination, with n-type absoφtion layers.

0044Figure 29 shows the monolithic integration of HIT cooler devices, with CMOS devices, with Avalanche PIN and HIP devices. This configuration is suitable for Back-Side Illumination.

0045Figure 30 shows an example of CMOS image sensors with Back-Side Illumination, made on Thin-Film SOI substrates. The PIN and HIP devices made on the same substrate are monolithically integrated with the CMOS devices.

0046Figures 31 to 40 show an exemplary process flow for an implementation on Thin-Film SOI substrates for Back-Side Illumination.

0047Description of exemplary embodiments of the invention

0048The present invention uses a process architecture which enables the fabrication of bandgap and doping engineered photodiodes, monolithically integrated with state of the art CMOS transistors. The exemplary process flow uses a selective or non-selective epitaxial growth of the photodiode device layers on silicon-based or germanium-based substrates.

0049Epitaxial growth of PIN photodiode devices integrated with CMOS devices on a semiconductor substrate is disclosed in WO/EPOl/11817. One of the many advantages of fabricating PIN (or PN) photodiodes through the epitaxial growth process step on active area "windows" formed on the CMOS wafer, is the straightforward incoφoration of Siι._x-yGe_xC_y alloys and/or superlattices during the epitaxial growth thereby allowing the improvement of different device parameters, such as coefficient of absoφtion and bandgap engineering, with respect to similar devices made with pure silicon. Such possibility is not available to pn-junctions formed in the bulk of the CMOS substrate, as is the case with conventional CCDs and CMOS image sensors.

0050For example, the incoφoration of Siι._x-yGe_xC_y random alloys and/or superlattices allows the cutoff wavelength to be extended to the 1.55μm wavelength range, which is considered to be "eye safe" and is of high relevance for fiber optics communications. Also, for each wavelength from IR to UN, the coefficient of absoφtion of Ge is much larger than the coefficient of absoφtion of pure silicon. Therefore Siι._xGe_x and Siι._x._yGe_xC_y random alloys and/or superlattices benefit of increased coefficient of absoφtion with respect to pure silicon.

0051However, extending the detection into longer wavelengths, such as MWIR seems very difficult with optoelectronic transitions across the band-gap (band-to-band) of Siι._x._yGe_xC_y random alloys and/or superlattices. Even though it has been predicted that the bandgap Siι_-yC_y random alloys can be reduced to zero it does not seem, at this point in time, straightforward to implement in a CMOS production environment. LWIR seems totally out of the question for SiGeC random alloys, unless pure GeC or GeCSn alloys did produce extremely narrow bandgaps, because the carbon concentration in the Siι_-yC_y alloys predicted to have nearly zero bandgap is, at the moment, too difficult to obtain in a manufacturing environment.

0052Photo-detector architectures such as the Heterojunction Internal Photoemission (HIP) devices, rather than PIN photodiodes, can also be implemented with Sii-_x._yGβ_xC_y films, and can cover the 3-5μm (MWIR) and the 8-12μm (LWIR) ranges. That is because the physics of these two device types is very different, i HIP photo-detectors the optoelectronic transitions take place over a band offset (either conduction or valence) between two different materials, rather than from band-to-band within the same material. Therefore HIP devices are either p-i-p (hght absoφtion by free holes) or n-i-n (light absoφtion by free electrons).

0053The obvious brute force approach to the fabrication of PIN and HIP types of detectors would be to perform two separate SEG process steps, each forming the layers necessary for each type of photo- detector. However, by carefully designing the doping and heterojunction profiles to be formed during the epitaxial growth, selective or non-selective, the same epitaxial layers can be used for PIN and HIP devices.

0054PIN devices are bipolar and thus require both types of doping, while HIP devices are unipolar and require only one kind of doping. A single epitaxial growth of the device layers for PIN and HIP (eitiier p-i-p or n-i-n) photodiodes is possible if the growth starts on silicon surfaces with complementary doping. Therefore, one kind of device is made on n+ active areas, while the other land of device in p+ active areas. The epitaxially grown layers are either undoped or in-situ doped with profiles that are suitable for PIN and HIP devices.

0055Referring to Figure 1 there is shown a first combination of PIN and HIP devices, in which the PIN device is made on n+ active area. Consequently the HIP device is implemented with p-i-p layers on a p+ active area. The top films of the PIN and HIP devices are p+ doped. Figure 2 shows a combination of PIN and HIP devices, in which the PIN device is made on p+ active area. Consequently the HIP device is implemented with n-i-n layers on a n+ active area. The top films of the PIN and HIP devices are n+ doped.

0056An alternative approach to combine PIN & HIP devices can also be done with just a single epitaxial deposition step, but having one of the electrodes being formed through a Schottky junction, rather than doping, hi this alternative approach, the epitaxial deposition occurs on active-areas with the same type of doping. Figure 3 shows a combination of PIN and HIP devices, with in-situ p-type doping of the initial epitaxial layers, thus leading to p-i-metal HIP devices. Figure 4 shows a combination of PIN and HIP devices, with in-situ n-type doping of the initial epitaxial layers, thus leading to n-i-metal HIP devices.

0057The details of the epitaxial profiles for the PIN and HIP devices and the advantages in implementing them in the fabrication of light-sensing devices will be discussed below. It will be apparent that, with the appropriate choice of substrate materials, device layers and heterojunction engineering and process architecture, it becomes possible to fabricate silicon-based and germanium-based multi- spectral sensors that can deliver pixel density and cost of fabrication comparable to the state of the art CCDs and CMOS image sensors. The present invention can be implemented with epitaxially deposited films on the following substrates: Silicon Bulk, Thick-Film and Thin-Film Silicon-On- Lisulator (SOI), Germanium Bulk, Thick-Film and Thin-Film Germanium-On-Insulator (GeOI).

0058Geometry & Layout of PIN & HIP for operation in Avalanche & Geiger modes

0059Epitaxial growth of photodiode devices provides photo-detectors that are decoupled from the structural parameters of the CMOS devices, such as doping profiles, junction depths, gate oxide, etc. For this reason, that process architecture allows the photo-detectors to be optimized independently from the CMOS devices. In particular, there are two key aspects that are impossible to implement with standard CMOS image sensors: fabrication of heterojunctions for bandgap engineering and operation in the Avalanche and Geiger modes. With bandgap engineering it is possible to enhance the coefficient of absoφtion of the photo- absorbing regions for the visible range, as well as to improve light detection for wavelengths outside the visible range. The Avalanche and Geiger modes of operation require the active region of the PIN photodiodes to be operated under very high electric fields, near and above the breakdown voltage of the active areas respectively.

0060The PIN & HIP devices fabricated according to the present invention can also be operated under the required bias conditions necessary for the Avalanche and Geiger modes of operation. Each device and device type (PIN & HIP) can be independently biased and can be operated in "normal bias", Avalanche and Geiger modes.

0061Epitaxial growth as disclosed in WO 02/033755 produces "vertical devices", in which current flows peφendicularly to the epitaxial layers. The deposited layers are flat and fully cover each active area. In order to maximize the functional advantages of operating the PIN & HIP devices in Avalanche & Geiger modes, it is necessary to pay special attention to layout-related aspects of the devices. In particular, it is necessary to avoid spurious currents from being amplified by the Avalanche/Geiger effects. Therefore, in devices fabricated according to that process flow it is necessary to avoid/suppress as much as possible all leakage currents originated at the edges between the active layers and the lateral isolation used in advanced CMOS technology: Shallow Trench Isolation (STI).

0062Figure 5 to 8 show three different geometries with which it is possible to implement the same device (PIN or HIP). These geometries provide examples of alternative strategies to minimize the probability of charges originated by the thermal generation recombination processes being multiplied by the avalanche process. Those charges are generated at the interface between the active areas and the isolation regions, which typically consist of trenches filled with silicon-oxide.

0063In particular, Figure 5 shows device layers as grown by selective epitaxial growth (SEG) on a silicon bulk substrate. In this case the photodiode active areas are non-uniformly doped so that the electric field at the edges of the active areas is much weaker than at the center. Figure 6 shows similar strategy for device layers which are epitaxially deposited in a non-selective way. The non- selective epitaxial growth of the layers in which the avalanche process takes place over a non- uniformly doped surface has one important advantage: the layers in which avalanching takes place are not surrounded by the isolation trenches, and therefore the generation/recombination originated by the surface states is completely removed from the high electric field regions, where avalanching takes place. On the other hand, the epitaxially deposited layers are amoφhous and or poly-crystalline over the silicon-oxide and/or silicon-nitride areas. For this reason, there is a boundary between the single- crystal film deposited over the active region and the amoφhous/poly film deposited over the isolation area. This boundary and the amoφhous and/or poly crystalline films have many structural defects that act as centers of generation/recombination. However, it is still expected that this geometry/layout is better at suppressing noise generated by edge-related defects than the one depicted in Figure 5.

0064Figures 7 and 8 show alternative geometries where the active area onto which the epitaxial films are deposited is not a simple planar surface. The lateral edges of the active area can be exposed in a very straightforward manner, and without extra process complexity, or cost. This geometry can be easily achieved by prolonging or tuning the pre-epitaxial cleaning process so that it results in an over-etch of the silicon oxide providing isolation between active areas. This geometry solves the problem of edge defects supplying spurious carriers to the avalanching regions, can be achieved with selective and non-selective epitaxial depositions, and in both cases the epitaxial films grown on the edges of the active region are single-crystal. The frill encapsulation of the bottom electrode (n+ or p+ active regions) results in the complete suppression of leakage currents along the lateral edges of the active regions and epitaxial active layers. Simulations and experimental results will determine how deep the recess of the field oxide needs to be in order to suppress leakage currents. The geometry of the electric field between the top electrode and the bottom electrode is dependent on the layout of two steps: the doping of the bottom electrode, i.e., surface of the active area, and the patterning of the top electrode.

0065In the particular combination of geometry and layout shown in Figure 1, the doping of the bottom electrode (surface of active area) is uniformly doped, and the patterning of the top electrode is recessed with respect to the edges of the active area. It should be noticed that the non-doped film in which avalanching takes place is not recessed and therefore fully covers the edges of the bottom electrode. With this combination of structural geometry and layout, the electric field between the top and bottom electrodes is much weaker at the edges than it is at the center of the structure. Because of the unifoπn doping of the bottom and top electrodes, the electric field should be constant in most of the cross section of the device, becoming weaker only near the edges. This geometry of the electric field insures that the threshold for avalanching is never reached at the edges of the active area, while there is unifonn avalanching everywhere else.

0066In the combination of geometry and layout shown in Figure 8, the bottom electrode is non-unifonnly doped, both in the horizontal direction, with higher doping concentration at the edges than at the center of the active area, and in the vertical direction, with a recessed profile, meaning high doping concentration at the top surface, and decreasing doping concentration when moving away from that surface. This particular combination of geometry and layout results in an electric field that is strongest at the corners of the active region, with rapid weakening when moving away from those comers.

0067The configuration shown in Figure 8 still prevents edge defects to inject carriers into the avalanching region, but allows a "comer effect" that for the a given applied voltage results in an electric field that is much larger than for a "planar interface" between the top and bottom electrodes, such as those shown in Figure 5, 6 and 7. The stronger electric field originated by the "comer effect", provides the possibility of impact ionization and therefore avalanche processes at voltages significantly lower than those necessary for "planar interfaces", that is, without the "comer effect". It should be realized that this is essentially a geometry effect, and that the breakdown of the junction will happen at lower voltages, still through impact ionization, rather than with band-to-band tunneling. The non- uniform doping of the bottom electrode helps to concentrate the electric field at the comers of the active areas. The electron-hole pairs generated at the center of the top electrode, which also acts as the absorbing layer, feel a drift electric field towards the comers of the active region where they are accelerated and promote impact ionization, i.e., avalanche multiplication.

0068It is worth noting that Figures 3 and 4 already show epitaxial layers which have geometries conducive to the "comer effect". These comers are a natural consequence of a process flow in which the epitaxy is perfonned inside a window of a dielectric such as Si₃N , which is necessary to provide an etch stop for the patterning of the epitaxial layers. Furthermore, the layout of the mask used to pattern the epitaxial window may be such that the number of comers thus generated can be either maximized or minimized.

0069The geometries and layouts shown in Figures 5 to 8 can be implemented for PIN devices alone, HIP devices alone, or combined PIN and HIP devices.

0070Optimal profiles of avalanche regions

0071The avalanche region is the "middle region", between the top and bottom electrodes, of both PIN devices and HIP devices. Either the top or the bottom electrode will also be the absoφtion layer for both the PIN and the HIP devices.

0072Because the avalanche region is also grown epitaxially, it is possible to perform bandgap and doping engineering in order to optimize the performance of the avalanche process. Photo-generated carriers will be injected from the absoφtion layer, and therefore the avalanche profiles can be oriented according to the position of the absoφtion layer.

0073The avalanche process should increase the signal-to-noise ratio (SNR) of the photo-absorbed signal. Therefore, the amplification of the incoming signal should be as large as possible, while the noise associated with the amplification process should be as low as possible. At the same time, from circuit integration and power dissipation standpoints, high amplification or multiplication factors should be obtained with the smallest voltage possible.

0074These requirements can be met with the "staircase" profiles described by F. Capasso et al. (IEEE Trans. Elec. Dev., Vol. 30, No. 4, pp. 381-390, 1983), which effectively lead to single-carrier or unipolar avalanche processes, thereby resulting in minimal "excess noise". The judicious insertion of narrow bandgap regions with a "Type-I" band alignment, in a "host" wider bandgap material, enables the blocking of the transport of one type of carriers. At the same time, when crossing the heterojunction interfaces with discontinuities in the potential energy, the other type of carrier gains additional kinetic energy in virtue of the conversion of excess potential energy into kinetic energy.

0075For one-sided junctions the minimum voltage required for causing impact ionization is about 6.6 times the magnitude of the bandgap. Therefore, by positioning the narrow bandgap regions at points in which there is a high probability of impact ionization, the voltage necessary to cause avalanching is lowered with respect to the substrate material.

0076Because smaller bandgap materials have higher intrinsic carrier concentration, and thus higher thermal noise, it is more advantageous to minimize the size of the small bandgap regions. For that reason, the avalanche regions consist of alternating wide bandgap layers, in which carriers are accelerated, and small bandgap layers, in which accelerated carriers impact ionize.

0077For electron-only avalanching, the valence-band edge must have discontinuities that impede the drift-diffusion current of holes. For this purpose, the valence-band edge discontinuities need to present a potential barrier that is high enough so that the only possible current mechanism for holes is thermionic emission of over the heterojunction barrier, and there is a complete suppression of hole current by drift-diffusion mechanisms. At the same time, when crossing the heterojunction interfaces with discontinuities in the potential energy, electrons gain additional kinetic energy in virtue of the conversion of excess potential energy into kinetic energy. For holes-only avalanching, the conduction-band edge must have discontinuities that impede the drift-diffusion current of electrons. For this puφose, the conduction-band edge discontinuities need to present a potential barrier that is high enough so that the only possible current mechanism for electrons is thermionic emission of over the heterojunction barrier, and there is a complete suppression of electron current by drift-diffusion mechanisms. At the same time, when crossing the heterojunction interfaces with discontinuities in the potential energy, holes gain additional kinetic energy in virtue of the conversion of excess potential energy into kinetic energy.

0078It should be emphasized that the design and fabrication of the avalanche multiplication layers are independent of the physical mechanisms employed for photo-absoφtion. For that reason the avalanche layers described below can be used with both PIN and HIP devices described in this application.

0079The bandgap engineering of the avalanche layers leads to an asymmetry with respect to the point of injection of the photo-generated carriers. As it will be explained later, the avalanche region can operate equally well for electrons or holes, depending only on the direction of the electric field. If the point of injection of the carriers does not change, but the carrier type does, then the only change necessary is to reverse the polarity and adjust the magnitude of the electric field.

0080Avalanche regions on Silicon substrates

0081The implementation of "staircase" heterojunction profiles for the avalanche regions, is fairly easy to achieve with on Si <100> and Si <111> substrates, with SiGeC random alloys and/or Si_m-Ge_n superlattices, and/or (Siι.yC_y)_m-Ge_n superlattices, and/or (Siι_-yC_y)_m-(Geι._xC_x)_n superlattices strained to silicon.

0082The incoφoration of SiGe and SiGeC random alloys and/or superlattices in the impact-ionization regions has another important consequence: it increases the impact-ionization rate of holes, which depending of the Ge content can even smpass that of electrons. This characteristic is detrimental for low noise operation of avalanche processes in uniform regions. However, for avalanche regions with "staircase" heterojunction profiles this characteristic does not impact the noise behavior, and enables hole-only avalanche regions with the same, or even higher, efficiency than electron-only avalanches.

0083Consequently, by careful design of the avalanche region it is possible to arrive at a profile that provides the same avalanche efficiency, onset of avalanche, gain, and noise behavior, for electron- only avalanching or for hole-only avalanching. In that case, by applying the same voltage amplitude, of the suitable polarity, to devices injecting electrons and holes, would result in the same avalanche multiplication factor, and the same "excess noise" factor for both carrier types. Then the parameters of the avalanche region become a neutral factor in the choice of the doping type for the absoφtion layer.

0084The same kinds of alloys and superlattices mentioned for normal, "unstrained", silicon substrates can also be used with "strained silicon substrates", or on SiGe relaxed buffer layers. Naturally these possibilities also apply to Silicon-On-Insulator (SOI) and Strained-Silicon-On-Insulator (SSOI) <100> and <111> substrates.

0085Figures 9 and 10 show schematic representations of the ideal energy-band diagram of avalanche layers that under a certain bias provide electron-only avalanche multiplication, and under the reverse polarity bias provide hole-only avalanche multiplication, as long as both types of carriers are injected from the same direction into the avalanche layers.

0086Figures 11 and 12 show schematic representations of avalanche layers and associated energy-band diagrams, implemented with SiGeC random alloys, or (Siι__yCy)_m-(Geι_-xCχ)_n short-period strained- layer superlattices on Si substrates.

0087Avalanche regions on Germanium substrates

0088The implementation of "staircase" heterojunction profiles for the avalanche regions is also fairly easy to achieve with on Ge <100> and Ge <111> substrates. Using SiGe and SnGe alloys and superlattices, it is possible to constmct conduction-band and valence-band offsets. It is also engineer strain compensation, since Ge is smaller than Sn but bigger than Si. This is an analogous situation to silicon substrates in which Carbon, being smaller than Si, can be used to compensate the strain of Ge which is bigger than Si.

0089Because the band alignments of these alloys and superlattices, strained to Ge substrates are not quantitatively well known, the schematic energy-band diagrams shown in Figures 9 and 10 depict desirable qualitative alignments.

0090Once the accurate data about these alignments becomes publicly available, it will be possible to make quantitative energy-band diagrams similar to those shown for the SiGeC random alloys and superlattices on silicon substrates. Optimal profiles for absorption region of Avalanche PIN devices

0091Separate Absoφtion and Multiplication Avalanche Photodiodes (SAM-APDs) are asymmetric PIN diodes. The "I" region is the avalanche region, and was described above. The absoφtion region, which is also an electrode, is substantially more complex than the other electrode.

0092The surface at which light penetrates the SAM-APD, is the side where the absoφtion layer needs to be. This asymmetry is very important because it defines the proper layer sequence for the avalanche region and for the absoφtion region. For the absoφtion region this asymmetry is important because it defines the orientation of the built-in drift field that drives carriers into the avalanche region, which must have its band offsets oriented accordingly.

0093Ideally the absoφtion layer, which is simultaneously an electrode, has the following properties: high coefficient of absoφtion for all the wavelengths of interest, built-in drift field for driving photo- generated carriers into the avalanche region, and low electrical resistance.

0094On one hand the absoφtion region should be a thin-film so that there is a minimal distance for the photo-generated carriers to travel through, before reaching the avalanching region. On the other hand, the photo-absoφtion should be as complete as possible, for which thicker photo-absoφtion fihns are better. In any case the absoφtion should be as high as possible because it generates a stronger signal before avalanche amplification, and because it avoids photo-generation in other layers of the device, which in turn avoids the injection of two types of carriers into the avalanche region.

0095The avalanche layer, its requirements and its design are described earlier herein. The essential requirement for the "other" electrode layer is to be highly conductive.

0096Absorption region of PIN devices on Silicon Substrates

0097As mentioned above, this region should be a highly efficient light absorber, have a drift field for the photo-generated carriers, and have low resistance. It is well known that Germanium has much higher coefficients of absoφtion photo-absorber than Silicon, because of its 0.66eV (versus 1.12eV in Si) minimum bandgap, and because its smallest direct bandgap is 0.8eN (versus 3.4eV in Si).

0098It is also known that Siι-_xGe_x and/or Siι._x._yGe_xC_y random alloys strained to silicon <100> substrates retain the essential features of the band structure of silicon, that is, the bandgap is reduced with increased gennanium content mainly because of a shift in valence band edge. Similarly to silicon, the direct bandgaps of SiGe are much larger than the minimum indirect bandgap. However the alloys show appreciable improvements in the coefficients of absoφtion for the visible range and extend the wavelength of detection deeper into the SWIR range. There seems to be a fairly linear relation between the improvement in coefficient of absoφtion and the amount of Ge in the films.

0099Strained-layer superlattices (SLS) consisting of alternating four monolayers of pure Ge and four monolayers of Si, all strained to silicon <100> substrates, have been experimentally verified to have very enhanced optoelectronic properties. It is seen that the experimental conditions under which the "Electroreflectance" is measured, namely the externally applied electric field, these superlattices seemed to have a direct bandgap with a magnitude similar to the direct bandgap of pure Ge, that is around 0.76eV, which is even less than the energy of photons with the wavelength of 1.55μm (Energy = 0.8eV).

0100For <100> silicon substrates it seems that 4 is the maximum number of Ge monolayers possible to grow pseudomoφhically without relaxation. On <111> silicon substrates it seems that n=5 or larger should be possible, hi any case, 4 monolayers of pure Ge on <100> Si substrates is technologically very difficult to achieve. On the other hand, the larger the number of monolayers the less sensitive the band stmcture of the superlattice will be to monolayer fluctuation inherent to the epitaxial deposition process.

0101Theoretical calculations of the band structure of short-period strained-layer Ge_n-Si_m superlattices on <100> or <111> Si substrates show that the smallest bandgap is always indirect, regardless of the values for "n" and "m". In these superlattices the conduction-band edge is formed by bulk-Si Δ states, while the top of the valence-band originates in the bulk-Ge T-point states. The zone-folding of the Silicon conduction-band results in a direct bandgap material only if the 2-fold Δx minima are lower in energy than the 4-fold A_\\ ones. This effect can be achieved in two ways:

01021. The 4-fold Δ.u minima move up by putting silicon under tensile strain. That can be done with the epitaxial growth of silicon films on Ge substrates or SiGe relaxed buffers. 2. The 2-fold Δx minima move down by putting silicon under compressive strain. That can be done with the epitaxial growth of silicon doped with carbon atoms, forming Siι_-yC_y random alloys.

0103Growing epitaxial silicon on Ge substrates or SiGe relaxed buffer layers is not suitable for integration with CMOS, miless the CMOS devices are themselves made on strained silicon substrates. On the other hand, it is well documented, the incoφoration of Carbon atoms in silicon films induced the lowering of the 2-fold Δ minima with respect to the 4-fold Aj minima. This is precisely what is necessary for the zone-folding of the silicon conduction-band to result in a pseudo- direct bandgap.

0104Fully pseudomoφhic Ge_i-yCy random alloys on Si <100> substrates have been experimentally demonstrated by M. Todd et al. (Appl. Phys. Lett. 68, No. 17, 22 April 2407 1996, pp. 2407-2409). Even though the band alignments with respect to silicon are not characterized as a function of Carbon content, it is reasonable to expect that for fairly small amounts of carbon, the band structure will be that of pure Ge strained to silicon, with lowering of the conduction-band, similarly to Siι_-yC_y random alloys. The most important advantage of using Geι__xC_x random alloys strained to <100> Si, instead of pure Ge strained to <100> Si, is that rather than being limited to a maximum of four monloayers, much thicker films can be grown due to the strain compensation provided by Carbon, hi turn, this enables the fabrication of Si_m-(Geι_-xC_x)_n or (Siι-yC_y)_m-(Geι._xC_x)„ superlattices with much larger periods, i.e., with "m" and "n" much larger than 4 or 5, which has beneficial consequences for the bandgap engineering.

0105It is hereby disclosed a new strained-layer superlattice on <100> silicon substrates, that from a theoretical standpoint should induce the zone-folding of the conduction-band of silicon into a pseudo-direct bandgap material. The new superlattice is: (Si)-yCy)_m-(Geι_-xC_x)_n, that is, "m" monolayers of Siι._yC_y random alloy and "n" monolayers of Geι._yC_y random alloy, all strained to silicon <100> or <111> substrates.

0106The magnitude of the lowering of the 2-fold Δ minima, with respect to the 4-fold A_\\ minima, depends on the percentage of carbon present in the alloy. Therefore, it is possible to do extra bandgap engineering by playing with the carbon composition across the total thickness of the superlattice. One possible way to do this, is to have several superlattices, each with consisting of a number of periods of (Siι._yC_y)_nι-(Geι__xC_x)_n, and each having a different percentage of Carbon atoms in the (Siι.._yC_y) random alloy. This way, it is possible to engineer abrupt heterojunctions or smooth variations of the conduction-band edge across the entire superlattice, and thus create a built-in electric drift field across the entire superlattice.

0107One more, and very important, benefit of the incoφoration of carbon in the superlattice is the compensation (at least partial) of the strain induced by the presence of the 4 monolayers of pure Ge. The amount of strain compensation depends on the amount of Carbon in the films. Strain compensation can be used to increase the total thickness (number of periods) of the superlattice, far beyond what is the critical thickness for the superlattice without Carbon atoms.

0108The ideal absoφtion layer of a SAM-APD on Silicon <100> substrates is thus composed of multiple (Sii._yCy)_m-(Gei-_xC_x)_n, periods, each with its own carbon concentration in order to have a built-in field that drives the photo-generated carriers into the avalanche region.

0109The absoφtion layers with the heterojunction and doping profiles described in this section can be implemented with any of the geometries and layout features shown in Figures 5 to 8.

0110Figures 13 and 14 show schematics of the ideal heterojunction and doping profiles, along with schematics of the corresponding energy-band diagrams, of PIN SAM-APDs for Si substrates using SiGeC random alloys. In particular, Figure 9 depicts a SAM-APD with p-type doped top- electrode/absoφtion-layer for electron injection and avalanching, and Figure 10 depicts a SAM- APD with n-type doped top-electrode/absoφtion-layer for hole injection avalanching.

0111Figures 15 and 16 show schematics of the ideal heterojunction and doping profiles, along with schematics of the corresponding energy-band diagrams, of PIN SAM-APDs for Si substrates using (Siι-yCy)_m-(Siι_-x-yGe_xC_y)_n short-period strained-layer superlattices. Figure 15 depicts a SAM-APD with p-type doped top-electrode/absoφtion-layer for electron injection and avalanching. Figure 16 depicts a SAM-APD with n-type doped top-electrode/absoφtion-layer for hole injection and avalanching.

0112Absorption region of PIN devices on Germanium Substrates

0113Theoretical calculations show that Si_mGe„ superlattices strained to Ge <100> and <111> substrates can have a pseudo-direct bandgap. Several combinations of "m" and "n" result in pseudo-direct bandgaps with slightly different characteristics. In addition to enabling pseudo-direct bandgaps, SiGe and SiGeC alloys and/or superlattices can be useful to perform bandgap engineering for puφoses such as to construct a built-in drift field across the superlattice, and to construct an avalanche region in which only unipolar avalanche processes take place.

0114Germanium substrates are suitable for other types of Group IN alloys and/or superlattices, such as those including Sn (Tin): GeSn, SiGeSn, SiGeSnC, SiSnC, SnC, strained to the germanium substrate. GeSn random alloys and or superlattices with the right stoichiometry have direct bandgaps. The epitaxial growth of such alloys has been demonstrated on Ge substrates, with <100> orientation. Given that the conduction-band minima of Ge occur along the <111> directions, and given that the transversal electron mobility in Ge is highest on <111> substrates, these substrates are the preferred ones for electronic (such as MOSFETs) and optoelectronic devices (such as APDs) using band-to-band transitions. The incoφoration of Si and C atoms in the GeSn alloys can be used to construct conduction- and/or valence-band edge discontinuities, as well as to provide strain- compensation to the Sn atoms, which are much larger than the Ge atoms.

0115Because the band alignments of these alloys and superlattices, strained to Ge substrates are not well known, the schematic energy-band diagrams of Figures 13 to 16 depict the desirable qualitative alignments. Once the accurate data about these ahgnments becomes publicly available, it will be possible to make quantitative energy-band diagrams similar to those shown for the SiGeC random alloys and superlattices on silicon substrates.

0116Device layer fabrication for Front-Side Illumination

0117"Front-Side Illumination" refers to a situation in which light penetrates the light-sensing layers having traveled through the hater-Metal Dielectric layers. This illumination condition is the most common one, and can be implemented by fabricating the devices on bulk Si or Ge substrates, thick- film SOI, thick-film GeOI, or Thin-Film SOI, or Thin-Film GeOI. With this illumination condition, the top electrode is also the absoφtion layer.

0118Figure 1 shows CMOS devices with the Avalanche PIN photodiode having a p-type absoφtion layer and top electrode. This configuration requires the Avalanche PIN photodiode to have a n-type bottom electrode.

0119Figure 2 shows CMOS devices with the Avalanche PIN photodiode having a n-type absoφtion layer and top electrode. This configuration requires the Avalanche PIN photodiode to have a p-type bottom electrode.

0120Device layer fabrication for Back-Side Illumination

0121"Back-Side Illumination" refers to a situation in which light penetrates the light-sensing layers from the side opposite to that having the multi-level metal interconnects. The simplest way to implement this illumination condition is to fabricate the devices on Thin-Film SOI substrates, or Thin-Film GeOI substrates. With these substrates, the bottom electrode is also the absoφtion layer. In Figure 3 there is shown CMOS devices with the Avalanche PIN photodiode having a p-type absoφtion layer and bottom electrode, and having a Schottky junction top electrode. Figure 4 shows CMOS devices with the Avalanche PIN photodiode having a n-type absoφtion layer and bottom electrode, and having a Schottky junction top electrode.

0122The same final arrangement, that is, absoφtion layer facing "down", can be achieved through a different route, in which the epitaxial deposition of the device layers is performed on a sacrificial substrate. After the epitaxial deposition, the device layers are separated from the sacrificial substrate and then transferred to the final substrate, which is transparent to light. The attachment of the epitaxial device layers to the final substrate is such that the last layers to be grown are the ones at the interface with the final, light-transparent, substrate. Naturally the separation of the device layers from the sacrificial substrate, as well as the attachment to the final substrate, are done at temperatures low enough to avoid changes in the as-grown heterojunction and doping profiles.

0123Profiles for absorption region of Avalanche HIP devices

0124Infra-Red (TR) radiation with photon energies smaller than the bandgap of the materials in the PIN photodiodes, cannot be sensed by the PIN photodiodes. For the range of wavelengths in the 3μm to 5μm and 8μm to 12μm, sensing can be performed by free-carrier photo-absoφtion, with the cutoff wavelength being defined by the barrier height of a semiconductor-semiconductor heterojunction (Heterojunction Internal Photoemission - HIP devices), or by the barrier height of a metal- semiconductor heterojunction (Schottky Photodiodes).

0125In the silicon materials system HIP devices can be easily implemented with p-Si/p-Siι_-xGe_x heterojunctions, in which the barrier height is defined by the percentage of germanium in the random alloys and by the p-type doping level and profile across the Si Siι__xGe_x heterojimction.

0126Si/SiGe HIP devices are preferable to Schottky type devices because it is possible to control the SiGe alloy stoichiometry, the doping level and profiles with a high degree of accuracy, and thus fine-tune the barrier height as well as other parameters. Schottky barrier devices can be made only with a limited set of materials, metals and metal-silicides, and therefore it is not possible to continuously vary the photo-absoφtion and/or transport properties.

0127State of the Art SiGe/Si HIP photo-detectors are not suitable for room temperature operation. Conceptually the HIP devices can operate at room temperature, but the implementations of the concept with SiGe/Si materials have, thus far, been unsuccessful. There are several physical parameters controlling the efficiency of the SiGe/Si HIP devices that in those implementations compete against each other, and the final performance is not good enough for room temperature operation. In the following, a new implementation of the HIP device concept with SiGeC materials provides a solution to decouple some of the parameters limiting the performance of conventional SiGe/Si HIP devices, thus providing a path for significant improvements in performance that will allow the operating temperature to be increased.

0128The physical modeling of HIP devices provides the insight about which parameters are control the performance of HEP detectors. In qualitative terms it can be said that the performance of the HIP devices is determined by:

0129a) Free carrier concentration in the absorbing layer. Typically, the free carrier concentration can be increased by increasing the doping concentration in the absorbing layer. b) Lifetimes of the carriers in the absorbing layer. The main factor for the decrease of the lifetime has been determined to be the large doping impurity concentration required to provide the free carriers. c) Free carrier mobility in the plane peφendicular to the epitaxial deposition, which typically is also peφendicular to incident light. Mobility is dependent on the lifetime and on the effective mass of charge carriers. Given that the lifetime is largely determined by the doping concentration, an increase in mobility could be obtained by lowering the hole effective mass, that is, by increasing the concentration of Ge in the absorbing film. However, that leads to an increase of the heterojunction barrier height with respect to silicon, and therefore to a decrease of the cutoff wavelengtli.

0130hi addition to improvements in the absoφtion capabilities, through more sophisticated heterojunction and doping profiles, than currently available as State of the Art, the absoφtion layers are coupled to the avalanche regions described earlier, for high gain and low noise operation. This results in Separated Absoφtion and Multiplication (SAM) Avalanche HIP devices. The avalanche multiplication could provide the performance increase necessary for these devices to have acceptable operation at room temperature.

0131Four new device layer profiles are hereby disclosed, that should lead to a considerable improvement in the performance of SiGe-based HIP devices, possibly leading to good enough performance for room temperature operation: p-type HIP, having heterojunction formed by two different SiGeC random alloys n-type HIP, having heterojunction formed by two different SiGeC random alloy p-type HIP, having heterojunction formed by two different (Siι_-yC_y)₄-Ge superlattices n-type HIP, having heterojimction formed by two different (Siι._yC_y) -Ge superlattices

0132p-tvpe HIP with SiGeC random alloys

0133In prior art, the heterojunction determining the cutoff wavelength is always between pure silicon and SiGe, hence the name Si/SiGe HIP devices (or SiGe/Si or GeSi/Si, etc.). For a fixed target wavelength of operation, this immediately puts rigid constraints on how much Ge and how much doping needs to be in the absorbing layer.

0134The key to break the three constraints listed above is to make the heterojunction of the HIP device that controls the cutoff wavelength, with different parameters than what has been done in prior art. It is hereby proposed to make SiGeC/SiGeC p-type HIP devices, in which the band offset controlling the cutoff wavelength is engineered on both sides of the heterojunction. The incoφoration of Carbon allows for partial or total strain compensation of the Ge atoms, and thus enables the fabrication of heterojunctions with higher Ge content and/or much thicker films. For a given cutoff wavelength, and thus for a given "barrier height", the incoφoration of Ge and C on both sides of the heterojunction provides more flexibility on how to achieve that particular barrier height. For example it allows for higher Ge content and lower doping on the absoφtion layer, both leading to higher hole mobility and thus to higher detectivity. The new heterojunction design is schematically shown in Figure 17.

0135hi addition to decreasing the hole effective mass and increasing hole mobility, it is also possible to create complex SiGeC profiles (rather than just a constant film composition) inside the absorbing layer itself, such that some regions can be lowly doped, but that are nonetheless populated with very high free carrier concentration due to effects of "carrier spilling" from adjacent layers at room temperature, effectively achieving a sort of "modulation doping" without creating quantum wells. The expectation is that the free carriers that are "spilled over", enjoy very long lifetimes, and thus further enhancing the mobility.

0136n-type HIP with SiGeC random alloys

0137The reason why Si/SiGe HIP devices have been only p-type devices is because SiGe alloys can provide large valence-band discontinuities, and because conventional epitaxial deposition techniques, such as MBE and CND, have shown very good control in tailoring Boron profiles with in-situ. At the same time, the conduction-band offset obtainable with Si/SiGe heterojunctions strained to <100> silicon is negligible, and shaφ n-type doping profiles have been notoriously difficult to achieve with in-situ doping. For these reasons n-type HIP devices on <100> silicon substrates did not seem feasible.

0138The demonstration of high-quality Siι_-yC_y and Siι_-x-_yGe_xC_y random alloys made possible the fabrication of conduction-band offsets with increased electron mobility, at least for very moderate amounts of carbon. It has since been verified that carbon, in addition to compensating the strain due to Ge, also suppresses Boron diffusion, and more recently has also been shown to be extremely helpful in achieving shaφ doping profiles during in-situ phosphorous doping with Rapid Thermal CVD.

0139With high quality Siι._yC_y and Siι-_x._yGe_xC_y random alloys, and abmpt n-type doping profiles, n-type silicon-based HIP devices have the potential to be superior to the p-type Si/SiGe HIP devices of prior art. Some of the best results with p-type Si/Sii._xGe_x HIP devices show that hole mobility values in the range of 28 cm²/V-s) to 44 cm²/V-s). These values are much lower than the hole mobility of undoped silicon (450 cm²/V-s), which can be taken as a reference. Assuming that n-type Si/Siι_-yC_y HIP devices will have an electron mobility that is also just about 10% of the electron mobility of undoped silicon (1500 cm²/V-s), that is still about 300% better than some of the best results with p-type Si/Siι_-xGeχ HIP devices.

0140Since the starting point, room temperature electron mobility in undoped silicon (1500 cm² V-s), and strained Si_]-yC_y and Siι_-x._yGe_xC_y random alloys, is much larger than the hole mobility in undoped Si layers (450 cm /V-s), it is therefore reasonable to expect that n-type Si/Si_1-yC_y HIP devices can perform better and at higher temperature than p-type Si/Siι_-xGe_x HIP devices.

0141There are two main technological solutions to further increase the electron mobility in sihcon: One is with Si/Siι-_yC_y or Si/Siι_-x._yGeχC_y layers, which in addition to lowering the conduction band edge (Ec), should also increase the electron mobility due to the lifting of the 6-fold degeneracy of the conduction band, with the consequent reduction in inter-valley phonon scattering.

0142The other has akeady been experimentally demonstrated: increase in room temperature electron mobility, by straining silicon on SiGe relaxed buffer layers. For example, for SiGe buffer layer with 20% Ge, the electron mobility of silicon increases by a factor of 1.8. Buffers with higher percentage of Ge provide even larger enhancements of the electron mobility in the silicon layer. Naturally, with higher strain there is a reduction in the critical thickness of the strained (silicon) layer. Ultimately it might be possible to reach or even surpass the room temperature electron mobility of Germanium:

01433900cmW

0144Strained Silicon on Insulator (SSOI) seems to be in the roadmap for a number of large semiconductor manufacturers, for the 65nm CMOS generation and below. Therefore, such type of substrates (SSOI) is likely to become widely available and possibly become the standard substrate for high performance.

0145The following is a simple calculation of performance ratio between of SiGeC/Si n-type HIP on SSOI substrate at 300K and a conventional SiGe/Si p-type HIP on bulk silicon substrates.

0146The dark current is given by:

0147J_ dark = A^**TV^Ψ/K°^T

0148Increasing temperature from 77K to 300K increases dark current by:

0149<img file="WO2004027879A2_D0001.tif" />

0150The variation in Detectivity is given by:

0151P^*(30CK) _cc Abs(300K) I J_dark(77K) _ 6 _{„ Q 22} D^*(77K) Abs(77K) ^" | J_dark(300K) 747

0152The effective Richardson constant A^{* *} is different for p-type and n-type Schottky junctions on silicon. For <100> silicon:

0153<img file="WO2004027879A2_D0002.tif" />

0154However, it should be noticed that A^*1 depends on the effective mass of the semiconductor, and on the carrier mean free path. Thus, if rather than pure silicon, one has SiGeC, with a significantly different electron effective mass, and a reduced intervalley scattering due to the lifting in the degeneracy in the conduction band (with consequent increase in mean free path), one is then lead to conclude that the effective Richardson constant can also be engineered for optimal performance. In conclusion, SiGeC/Si n-type HIP devices on SSOI substrate with 20%Ge, could have a performance (detectivity) at 300K that is about 22% that of SiGe/Si p-type HIP devices working at 77K. Even better performance can be attained with higher electron mobility, which can be accomplished with increased strain in the "strained Si" layer, and/or through the incoφoration of carbon. These calculations do not include the improvement in Signal-to-Noise Ratio (SNR) of the avalanche multiplication. Once that is also taken into account, it is reasonable to expect that n-type Avalanche HIP devices, even on unstrained silicon substrates, operating at room temperature, can far outperform conventional p-type Si/SiGe HIP devices operating at 77 .

0155Figure 18 shows an exemplary profile and related energy-band diagram for n-type Si/Siι__yC_y HIP devices.

0156p-type HIP with SiGeC superlattices

0157As stated above, for (Siι_-yCy)_πι-(Geι._xC_x)_rl superlattices, the conduction-band of the superlattice is mainly derived from the conduction band of the silicon-carbon random alloy, while the valence band of the superlattice is mainly derived from the valence-band of germanium. Accordingly, the ' fabrication of stacks of superlattices with different carbon contents enables the engineering of the conduction-band edge along the superlattice. Naturally it is possible to have a smooth variation of the valence-band edge, as well as ab pt heterojunctions if pure Ge is replaced with a Siι_-xGeχ (with Ge content larger than 70%) random alloy, resulting in a (Siι._yCy)_m-(Siι._x._yGe_xCy)_n superlattice.

0158Therefore, it is possible to engineer a p-type HIP device on silicon having the valence-band discontinuity formed by (Siι_-yC_y)m-(Siι_-x-yGe_xCy)n superlattices on both sides of the discontinuity. As mentioned above, for <100> silicon substrates, the maximum number of monolayers of pure Ge before relaxation is four. For silicon <111>, the maximum number of monolayers of pure Ge is above five. On <100> silicon substrates, for Siι._x._yGe_xC_y rather than pure Ge, the number of monolayers possible without relaxation is certainly above four, depending on the Ge and Carbon concentrations. In Ge bulk substrates, the valence-band is degenerate. In SiGe random alloys and pure Ge strained to Si substrates, the edge of the valence-band is provided by light-hole band alone. Therefore, even though the hole-effective masses do not change much from the pure elements (Si and Ge) to the superlattices, in the superlattices electrical transport is performed by light-holes only. Consequently, there should be an increase in mobility due to the reduction in mass.

0159Figure 19 the schematic of the device layers and the corresponding energy-band diagram. n-type HIP with SiGeC superlattices

0160As stated above, for (Siι-yC_y)_m-(Geι._xC_x)n superlattices, the conduction-band of the superlattice is mainly derived from the conduction band of the silicon-carbon random alloy, while the valence band of the superlattice is mainly derived from the valence-band of germanium. Accordingly, the fabrication of stacks of superlattices with different carbon contents enables the engineering of the conduction-band edge along the superlattice. Naturally it is possible to have a smooth variation of the valence-band edge, as well as abmpt heterojunctions if pure Ge is replaced with a Si_1-xGe_x (with Ge content larger than 70%) random alloy, resulting in a (Siι._yC_y)_ιn-(Siι_-x-yGe_xCy)_n superlattice.

0161Therefore, it is possible to engineer a n-type HIP device on silicon <100> and <111> substrates having the conduction-band discontinuity foπned by (Siι._yC_y)_m-(Geι_-xC_x)_n superlattices on both sides of the discontinuity. The conduction-band discontinuity between superlattices will be determined by the amount of Carbon present in each of the superlattices.

0162Theoretical calculations for Si_m-Ge_n superlattice grown on silicon substrates show that the electron effective mass in the superlattice is very similar to the silicon bulk values. It is then reasonable to assume that for (Siι_-yCy)_m-(Geι__xC_x)_n superlattices the electron effective mass in the superlattice will be very similar to the electron effective mass in Siι_-yC_y random alloys, which for moderate amounts of carbon is lower than in pure silicon. Figure 20 the schematic of the device layers and the corresponding energy-band diagram.

0163Absorption region of Avalanche HIP devices on Germanium Substrates

0164HEP devices built on Ge bulk substrates can take advantage of the extiemely high electron mobility values to shaφly increase the free-carrier absoφtion and thus the detectivity and signal-to-noise ratio. As a reminder, the electron mobility in bulk Ge is about 3900 cm²/V-s, while in bulk silicon is only 1500 cm²/V-s. This is one of the driving forces behind a renewed interest on Ge substrates for advanced CMOS technology.

0165Therefore, Ge-based n-type HIP devices are extiemely promising for MWIR and LWIR. Electron mobility is a function of effective electron mass: the lower the effective mass, the higher the mobility.

0166For Ge <111> substrates, the longitudinal mass is 1.64m₀ and the transversal mass is 0.082rno. Another extremely important advantage of germanium substrates is the feasibility of fabricating short-period strained-layer superlattices incoφorating Gray-Tin (α-Sn). Theoretical work shows that several superlattice compositions have direct bandgaps, on <111> and also on <100> germanium substrates.

0167Strained to Ge <111> substrates, a direct bandgap can be obtained with several SnιGe_n superlattice configurations, including "n"=3, that is, SnιGe₃. Strained to Ge <100> substrates, a direct bandgap can be obtained with several superlattice Sn₂Ge_n configurations, including "n' -2, that is, Sn₂Geι₀. SnιGe₃ superlattice are easier to grow on <111> Ge substrates than Sn₂Ge_l0 to grow on Ge <100> substrates.

0168SnιGe₃ grown on Ge <111> substrates could be easier to fabricate and include in different types of electronic and optoelectronic devices. For this configuration the smallest bandgap is about 0.22eV, which is small enough for the detection of MWIR through band-to-band transitions. Therefore, only LWIR and/or VLWIR sensing requires the fabrication of HIP devices.

0169The hole effective masses for bulk Ge are <img file="WO2004027879A2_D0003.tif" /> These bands are degenerate and therefore the advantages of the very low light-hole mass cannot be fully taken advantage of. For the SnjGe₃ superlattices on Ge <111> substrates, in the plane peφendicular to the direction of epitaxial growth, the two lowest hole-subbands have masses of only 0.078m₀, and 0.072m₀, and these two subbbands are not degenerate, thereby suppressing scattering.

0170The electron effective mass for bulk Ge is 0.082m₀, with 8 degenerate valleys. For the SnιGe₃ superlattices on Ge <111> substrates, the minimum of the conduction-band is at the T-point (direct bandgap), and the electron mass is 0.046m₀, in the plane peφendicular to the direction of epitaxial growth, and 0.053m₀ in the direction parallel to the epitaxial growth.

0171Therefore, it can be concluded that the fabrication of SnιGe₃ superlattices on Ge <111> substrates are not only advantageous for devices based on band-to-band transitions across a very small (0.22eV) direct bandgap, but also for devices based on free carrier absoφtion, both in the valence- band and conduction-band. The 0.22eV gap is not a pseudo-direct bandgap formed by zone-folding, but it is a truly direct b andgap .

0172Because the band ahgnments of these alloys and superlattices, strained to Ge substrates are not quantitatively well known, the schematic energy-band diagrams of Figures 17, 18, 19, and 20, depict desirable qualitative ahgnments. Once the accurate data about these alignments becomes publicly available, it will be possible to make quantitative energy-band diagrams similar to those shown for the SiGeC random alloys and superlattices on silicon substrates.

0173Front-Side Illumination versus Back-Side Illumination

0174For silicon substrates, all the alloys and superlattices have bandgaps such that band-to-band transitions are impossible for photons with wavelengths longer than SWIR. For this reason, the question of Front-Side and Back-Side illumination for MWTR and LWIR is irrelevant. Therefore the HIP heterojunction can be formed near the top electrode or the bottom electrode.

0175For germanium substrates, very small direct bandgaps can be made. For example, the SnιGe₃ superlattice has a direct bandgap of 0.22eV, which is small enough for a PIN device to cover the entire MWIR range. Therefore, the question of Front-Side versus Back-Side Illumination for such particular device and wavelength is relevant because the photo-absoφtion and carrier injection could take place from the undesirable side of the avalanche region.

0176HIP devices on Ge substrates for LWTR and VLWIR, are not affected by the issues of Front-Side versus Back-Side Illumination.

0177Profiles for combined PIN &amp; HIP devices

0178Previous sections of the present disclosure, described the optimized profiles for PIN devices implemented on silicon or germanium substrates, separately from the optimized profiles for HIP devices implemented on silicon or germanium substrates. For both devices, there were profiles for Front-Side or Back-Side Illumination configurations.

0179In the present section it will be described how the device layer requirements for optimal performance of PIN and HIP devices can be met simultaneously. The combined requirements result in heterojunction and doping profiles, suitable to be implemented on sihcon or germanium substrates, for Front-Side or Back-Side Illumination configurations.

0180Both device types will be coupled to avalanche regions, thereby resulting in Separate Absoφtion and Multiplication (SAM) regions for Avalanche PIN devices and Avalanche HIP devices. The absoφtion regions for both devices are made on top of identical avalanche layers. The physical mechanism for the photo-absoφtion of PIN and HIP devices are quite different, but their absoφtion layers are engineered together so that a single set of epitaxial layers can function for both absoφtion mechanisms, depending only on the applied voltage.

0181The HEP device will be n-type ifthe PIN absoφtion layer is p-type doped, and it will be p-type ifthe PEST absoφtion layer is n-type doped. These combinations of PIN &amp; HE? devices also mean that for each combination both devices inject the same type of carrier into the avalanche region.

0182Regarding the absoφtion region of the PIN device, the most important parameters are the magnitude and nature (direct or indirect) of the bandgap, as well as the existence of a drift field (built-in or not) to drive the photo-generated carriers towards the avalanche region. For p-type absoφtion regions, the photo-generated carriers are electrons, and therefore the drift field for electrons is made with a slope in the conduction-band edge. For n-type absoφtion regions, the photo-generated carriers are holes, and therefore the drift field for holes is made with a slope in the valence-band edge. Building slopes in the conduction-band or valence-band of the absoφtion layer of a PIN device can be achieved through the grading of the bandgap, through graded alloy compositions, and also through the grading of the doping profile.

0183Because the HIP device operates through free carrier absoφtion, its heterojunction needs to be placed rather closely to a layer with high free carrier concentration, such as a highly doped layer or an undoped layer populated through modulation doping.

0184For the propose of combining a HIP device with a PIN device, the heterojunction of the HIP device is placed at the interface between the avalanche region and the absoφtion region of the PIN device.

0185When the PIN device has a p-type absoφtion layer, then the HIP device will be p-type and its heterojunction is formed in the valence-band. When the PIN device has a n-type absoφtion layer, then the HEP device will be n-type and its heterojunction is formed in the conduction-band.

0186The constmction of the heterojunction for the HIP device is done only in the same band (conduction or valence) that is highly doped for the PIN device, while the "other" band edge can be smooth. For that reason the insertion of the heterojunction for the HIP device does not interfere with the injection of photo-generated minority carriers from the absoφtion layer of the PIN device into the avalanche region. Front-Side Illumination on Si Bulk substrates

0187PIN on n+ active area, having p-type absoφtion layer (p-i-p HIP device)

0188Starting from the interface with the substrate, the films to be deposited by the epitaxial process are:

01891. Avalanche layers, as already previously defined in the present disclosure.

01902. Undoped graded Siι_-xGeχ or Sii-x.yGexCy.

0191The thickness of this film is expected to be around 5nm to lOnm. The puφose of this .film is to enable the engineering of the valence-band offset (barrier height for holes) between the previously grown undoped sihcon film, part of the avalanche layers, and the next film to be grown: p+ Si_ι._x-yGe_xCy. The presence of this film is entirely due to the need to reconcile conflicting requirements for the doping and heterojunction profiles of the p+ Siι_-x._yGe_xCy layer to be grown next.

01923. In-situ p-type doped absoφtion layer and top electrode for both PIN and HIP: p+ Sii_-x._yGe_xCy.

0193Figures 21 and 22 show the energy-band diagrams of the PIN &amp; HIP (p-i-p) devices side-by-side, monolithically integrated with CMOS devices. The epitaxial layer stack was designed for Front- Side

0194Illumination, with p-type absoφtion layers, hi particular, Figure 21 shows the energy-band diagram for the PIN device under suitable bias condition for operation in avalanche mode. The PIN device injects electrons into the avalanche region. Figure 22 shows the energy-band diagram for the HIP device under suitable bias condition for operation in avalanche mode. The HIP device injects .holes into the avalanche region. It can be seen that the heterojunction for the HIP device is placed right at the interface between the avalanche layers and the top electrode, that is, the p-type doped absoφtion layer of the PIN device. If the HIP device was not to be made, that heterojunction would not be fonned. But even though it is there, it does impact the operation of the PIN device.

0195The requirements for PIN and HIP device layers conflict when wanting to increase the absoφtion coefficient and wavelength range of the absorbing layer for the PIN device, and wanting a HIP device capable of detecting photons in the 8-12μm range. The conflict arises because for PIN devices the increase in the coefficient of absoφtion and detection of longer wavelengths require smaller band-gaps, which if accomplished mainly through the incoφoration of more germanium in the SiGeC film, leads to larger valence-band offset with silicon. That would lead to shorter wavelength cutoff of the HIP detector, because increasing the p+ doping of that layer might not be able to compensate the increase and valence-band offset. Shifting the position of the quasi-fermi level for holes inside the valence band of the p+ Siι._x._yGe_xC_y to a deeper level (more degenerate) can compensate an increase in the valence-band offset. However, doping concentration can only be increased up to the limit of solid solubility of the dopant impurity in the Sii._x.yGβ_xCy material.

0196The growth of this fihn should proceed with progressively increased percentages of Ge and C atoms. The grading does not have to be linear or uniform, but it should provide smooth conduction- and valence-band edges until the next interface. In principle the incoφoration of this film should not affect the operation of PIN devices, as there should not be any significant impact (offset) on the conduction-band edge, between the previously grown undoped silicon film, and the next film to be grown: p+ Siι_-x._yGe_xCy.

0197PIN on p+ active area, having n-type absoφtion layer (n-i-n HIP device)

0198Starting from the interface with the substrate, the films to be deposited by the epitaxial process are:

01991. Avalanche layers, as already previously defined in the present disclosure.

02002. Undoped graded Siι._yC_y or Siι._x-yGe_xC_y.

0201The thickness of this film is expected to be around 5nm to lOnm. The puφose of this film is to enable the engineering of the conduction band offset (barrier height for electrons) between the previously grown undoped silicon film, part of the avalanche layers, and the next film to be grown: n+ Siι_-x._yGe_xC_y. The presence of this film is entirely due to the need to reconcile conflicting requirements for the doping and heterojunction profiles of the n+ Siι.χ_-yGe_xC_y layer to be grown next.

02023. In-situ n-type doped absoφtion layer and top electrode for both PIN and HIP: n+ Siι_{- -y}Ge_xC_y.

0203Figures 23. and 24 show the energy -band diagrams of the PBST &amp; HIP (n-i-n) devices side-by-side, monolithically integrated with CMOS devices. The epitaxial layer stack was designed for Front- Side Illumination, with n-type absoφtion layers, hi particular, Figure 23 shows the energy-band diagram for the PIN device under suitable bias condition for operation in avalanche mode. The PIN device injects electrons into the avalanche region. Figure 24 shows the energy-band diagram for the HIP device under suitable bias condition for operation in avalanche mode. The HEP device injects holes into the avalanche region. It can be seen that the heterojunction for the HEP device is placed right at the interface between the avalanche layers and the top electrode, that is, the n-type doped absoφtion layer of the PIN device. If there was no intention of making a HIP device, that heterojunction would not be formed. But even though it is there, it is not supposed to impact the operation of the PEN device.

0204The requirements for PIN and HIP device layers conflict when wanting to increase the absoφtion coefficient and wavelengtli range of the absorbing layer for the PIN device, and wanting a HIP device capable of detecting photons in the 8-12μm range. The conflict arises because for PIN devices the increase in the coefficient of absoφtion and detection of longer wavelengths require smaller band-gaps, which if accomplished mainly through the incoφoration of more carbon in the Siι_-x._yGeχCy fihn, leads to larger conduction-band offset with silicon. That would lead to a shorter wavelength cutoff of the HIP detector, because increasing the n+ doping of that layer might not be able to compensate the increase and conduction-band offset. Shifting the position of the quasi-feπni level for electrons inside the conduction band of the n+ Siι_-x._yGe_xC_y to a deeper level (more degenerate) can compensate an increase in the conduction-band offset. However, doping concentration can only be increased up to the limit of solid solubility of the dopant impurity in the Siι-χ-yGe_xCy material.

0205The growth of this film should proceed with progressively increased percentages of Ge and C atoms. The grading does not have to be linear or uniform, but it should provide smooth conduction- and valence-band edges until the next interface. In principle the incoφoration of this film should not affect the operation of PIN devices, as there should not be any significant impact (offset) on the valence-band edge, between the previously grown undoped silicon film, and the next film to be grown: n+ Siι_-x-yGeχC_y.

0206Back-Side Illumination on Thin-Film SOI substrates

0207Back-side illumination requires the absoφtion layer to be at the bottom of the epitaxial stack, because it is from there that light penetrates the sensor. Having to do the epitaxial growth of the absoφtion layer at the bottom of the stack for both PIN and HIP devices requires both active areas to have the same type of doping, and requires special care for the formation of the top electrode. Naturally a single epitaxial growth step cannot have simultaneous in-situ doping for n-type and p- type layers. Therefore, the formation of the top electrode requires more processing steps and alternative solutions:

02081. Epitaxial growth of undoped top electrode film, followed by masked ion-implantation for the p-type electrodes, and masked ion-implantation for the n-type electrodes. 2. Epitaxial growth of in-situ doped top electrode film, followed by masked ion- implantation with complementary doping type. This implantation step must frilly compensate and exceed the doping concentration introduced during the epitaxial growth.

02093. The epitaxial deposition stops after the growth of the absoφtion and avalanche layers. Two masked selective or non-selective epitaxial growth steps will growth separately, the in-situ doped, n-type and p-type electrode layers.

02104. The epitaxial deposition stops after the growth of the absoφtion and avalanche layers. The top electrode for both the PIN and HIP devices is a metal forming a Schottky junction on the undoped or slightly doped top epitaxial layer. A Schottky junction on undoped silicon can be viewed as a "neural" junction, because it is an equally good contact to extract electrons or holes. Because this Schottky junction is after the avalanche region, its role as a contact is only to extract carriers, electrons and/or holes. This Schottky junction should inject neither electrons nor holes into the avalanche region, and for that reason the barrier height should be nearly symmetric, i.e., the workfunction of the metal or metal-silicide forming the Schottky contact, should be near the midgap energy level of sihcon.

0211It should be kept in mind that the requirements regarding heterojimction and doping profiles of the top electrode are not critical, and are far less sophisticated than those of the bottom electrode which is simultaneously the absorbing layer for both he PIN and HIP devices. The role of the top electrode in the operation of the PIN and HIP devices is only that of a charge collector. The Schottky heterojunction, with workfunction near the midgap of silicon prevents injection of electrons and holes into the avalanche region for any possible bias condition.

0212The metal forming the Schottky junction will act as a mirror for light that crosses the absoφtion layer and avalanche layers without being absorbed. This effect is not expected to be detrimental to the operation of the Avalanche PEST devices because it is supposed that almost all light is absorbed in the bottom-electiode-absorbing layer. On the other hand, for the HIP device, the existence of this mirror is expected to double its absoφtion capability, because typically only a few percent are ab sorbed with one p ass through the absorbing layer.

0213Since the energy of the photons absorbed by the HIP device is far smaller than the bandgap of any of the layers in the absoφtion layer, as well as in the avalanche region, and smaller than the threshold for internal emission at the Schottky junction itself, there will be no carrier injection into the avalanche region other than the HIP heterojunction. The Schottky junction as top electrode of PIN and HIP devices for Back-Side Illumination systems made on SOI or GeOI substrates, with provide the best solution from process complexity and device performance standpoints. For that reason, this is the solution depicted in the figures below. With this solution the PIN device is no longer a bipolar device, because it will become a "p-i-metal" or "n-i-metal" device. The HIP devices will become either "p-i-metal" or "n-i-metal" devices.

0214PIN with p-tvpe absoφtion layer (p-i-p HIP device)

0215The active areas for the PIN and HIP devices are all p-type doped. Starting from the interface with the substrate, the films to be deposited by the epitaxial process are:

02161. In-situ p-type doped absoφtion layer and top electrode for both PIN and HIP: p+ Sii-_x._yGe_xCy.

02172. Undoped graded Siι._xGe_x or Si_1-x._yGeχC_y.

0218The thickness of this film is expected to be around 5nm to lOnm. The puφose of this film is to enable the engineering of the valence-band offset (barrier height for holes) between the previously grown p+ Siι._x._yGe_xC_y films, and the next film to be grown: undoped sihcon film, part of the avalanche layers. The presence of this film is entirely due to the need to reconcile conflicting requirements for the doping and heterojunction profiles of the p+ Siι.χ_-yGe_xC_y layer to be grown previously. 3. Avalanche layers as previously defined herein.

02194. Fonnation of a Schottky junction with a metal or a metal-silicide, having workfunction near the midgap value of silicon.

0220Figures 25 and 26 show the energy-band diagrams of the Schottky PIN &amp; HIP (p-i-metal) devices side-by-side, monolithically integrated with CMOS devices. The epitaxial layer stack was designed for Back-Side Illumination, with p-type absoφtion layers. In particular, Figure 25 shows the energy-band diagram for the P-I-Metal device under suitable bias condition for operation in avalanche mode. The P-I-M device injects electrons into the avalanche region. Figure 26 shows the energy-band diagram for the HD? (p-i-metal) device under suitable bias condition for operation in avalanche mode. The HIP device injects holes into the avalanche region. It can be seen that the heterojunction for the HIP device is placed right at the interface between the avalanche layers and the bottom electrode, that is, the p-type doped absoφtion layer of the PIN (P-I- Metal) device. If there was no intention of making a HIP device, that heterojunction would not be formed. But even though it is there, it is not supposed to impact the operation of the PIN (P-I-Metal) device. From Figures 25 and 26 it can be concluded that the exact same device can be operated as a PIN or a HIP device, depending only on the applied voltage.

0221PIN with n-type absoφtion layer (n-i-n HIP device)

0222Starting from the interface with the substrate, the films to be deposited by the epitaxial process are:

02231. In-situ n-type doped absoφtion layer and top electrode for both PIN and HIP: n+ Siι_-x._yGe_xCy.

02242. Undoped graded Siι_-yC_y or Siι_-x._yGe_xC_y. The thickness of this fihn is expected to be around 5nm to lOnm. The pmpose of this film is to enable the engineering of the conduction band offset (barrier height for electrons) between the previously grown n+ Siι._x._yGe_xCy films, and the next film to be grown: undoped silicon film, part of the avalanche layers. The presence of this film is entirely due to the need to reconcile conflicting requirements for the doping and heterojunction profiles of the n+ Siι-_x._yGe_xC_y layer to be grown previously.

02253. Avalanche layers as previously defined herein.

02264. Formation of a Schottky junction with a metal or a metal-silicide, having workfunction near the midgap value of silicon.

0227Figures 27 and 28 show the energy-band diagrams of the Schottky PIN &amp; HIP (n-i-metal) devices side-by-side, monolithically integrated with CMOS devices. The epitaxial layer stack was designed for Back-Side Illumination, with n-type absoφtion layers. In particular, Figure 27 shows the energy-band diagram for the N-I-Metal device under suitable bias condition for operation in avalanche mode. The N-I-Metal device injects holes into the avalanche region.

0228Figure 28 shows the energy-band diagram for the HIP (n-i-metal) device under suitable bias condition for operation in avalanche mode. The HIP device injects electrons into the avalanche region. It can be seen that the heterojunction for the HIP device is placed right at the interface between the avalanche layers and the bottom electrode, that is, the n-type doped absoφtion layer of the PIN (N-I-Metal) device. If there was no intention of making a HIP device, that heterojunction would not be fonned. But even though it is there, it is not supposed to impact the operation of the PIN (N-I-Metal) device.

0229From Figures 27 and 28 it can be concluded that the exact same device can be operated as a PIN or a HIP device, depending only on the applied voltage. Back-Side Illumination with "Device-Layer Transfer" from Si substrates

0230Since "Device-Layer Transfer" results in turning the epitaxially grown layers "up side down", the required epitaxial profiles are identical to the ones deposited for Front-Side Illumination on Si bulk substrates.

0231Ge Substrates

0232Because the band alignments of GeSn and GeSiSn alloys and superlattices, strained to Ge substrates are not quantitatively well known, the schematic energy-band diagrams of Figures 21 to 28 depict the desirable qualitative alignments. Once the accurate data about these alignments becomes publicly available, it will be possible to make quantitative energy-band diagrams similar to those shown for the SiGeC random alloys and superlattices on silicon substrates.

0233HIT cooler devices on top of PIN &amp; HIP devices

0234Back-side illumination enables increased functionality through further sophistication of the epitaxial device layers. As already pointed out in WO/EPOl/11817, the epitaxial deposition of the photo- detector layers, also enables, during the same epitaxial growth step, the formation of device layers for Heterojunction Integrated Thermionic (HIT) coolers. These devices provide a cooling effect of one of their sides (top or bottom) when under the suitable voltage conditions, and have been experimentally demonstrated with SiGeC/Si superlattices by X. Fan et al. (Appl. Phys. Lett., 78(11), pp. 1580, 12 March 2001), thus making them fully compatible with the silicon processing.

0235Therefore, the advantages of the process flow disclosed in WO/EPOl/11817 with respect to the monolithic integration of HIT coolers at the pixel level with the photo-detector layers, can be much improved for back-side illumination. In terms of process flow, the basic difference between front- side and back-side illumination lies with the profiles in the epitaxial layers, depending on which type of device is on top or at the bottom, and where the light comes from (top or bottom). In fact, it is more advantageous to have the HIT coolers monolithically integrated on substrates for back-side illumination, because the heat extraction will be performed from the top layer, i.e., through metal lines, rather than from the bottom through, i.e., through the bulk sihcon substrate, which is a bad heat conductor.

0236The fabrication of the HIT devices can be very straightforward and low cost, because the epitaxial layers for the HIT devices can be made with the same epitaxial growth run used to make the PIN and HIP photodiodes, without interruptions. In this case the bottom electrode of the HIT device is deposited after the undoped "buffer" layer is formed on top of the avalanche region.

0237This undoped "buffer" layer is necessary for the formation of the Schottky top electrode of PIN and HIP devices as explained earlier, and it also provides a separation between the avalanche region and the bottom electrode of the HIT device. The fabrication of the Schottky junction requires a patterning step in which the HIT layers are removed, with wet etching for example, from the areas in which the Schottky junction is to be formed, hi the undoped "buffer layer" it is straightforward to embedded a sacrificial etch-stop marking layer, which will be removed before forming the Schottky junction itself.

0238The metal forming the Schottky junction will simultaneously be the top electrical contact of the PIN and HEP photodiodes, and an "ohmic" contact to the highly doped bottom electrode of the HIT device layers.

0239Figure 29 shows the monolithic integration of HIT coolers with "Schottky PIN &amp; HIP" devices made on active regions with the same type of doping on substrates for Back-Side Illumination. ^'

0240HIT devices are unipolar, and can be implemented with heterojunctions in the valence-band or in the conduction-band, requiring only p-type doping or n-type doping respectively. Similar band- alignments can be engineered on germanium substrates, thus making possible the monolithic integration of PIN + HIP + HIT devices also on Ge substrates. From the point of view of silicon process technology, it is better to use n-type HIT coolers because the etching step to remove the HIT layers from the areas where the Schottky junctions are to be made, is very similar to the patterning of the n-type emitter layer in standard SiGe BiCMOS technology, which stops on the p-type SiGe base layer. Similarly, the marker layer embedded in the undoped "buffer" could also be a SiGe layer, having no electrical or optical function, being there just for the puφose of being an etch-stop layer.

0241The performance of HIT devices, i.e., the efficiency of cooling per unit of area, increases with decreasing area of the device. Therefore, the layout for the patterning of the HIT device layers can be such that for a certain total layer area (and thus total cooling power) is best achieved by having a number of devices in parallel, each with a small area, than a single device with a total area equaling the sum of all the areas of the small devices. SOI and GeOI Image Sensors

0242The method of fabrication disclosed in WO 02/033755 is applicable to any of the photo-detector device stmctures mentioned above, because it adds the active layers to the substrate, regardless if it is Silicon or Germanium bulk, SOI or GeOI. This is the reason why it enables the fabrication of CMOS image sensors on Thin-Film SOI or Thin-Film GeOI substrates, which have many advantages over bulk silicon for the fabrication of advanced CMOS devices and circuits, in terms of performance, radiation hardness, fabrication flow, etc. However, that fabrication method makes the photodiodes independent from, and not affected by, the details of the physics of the MOSFETs fabricated on such substrates.

0243One particular aspect of SOI (and GeOI) substrates that usually is not considered very relevant for the purely electrical CMOS circuits fabricated on SOI, is that the buried oxide in SOI wafers provides an excellent "marker" layer for the removal of the thick silicon substrate underneath the buried oxide. From a process technology standpoint, the presence of that "marker" layer makes it easy to replace the thick silicon substrate by a material that is transparent to visible and UV light. The replacement of silicon by a light-transparent material, such as quartz, sapphire, etc, can be done before or after the full processing of the front-side of the wafer.

0244Given the commercial availability of low-temperature processes for the removal and bonding to a new substrate, it is then preferable to perform the replacement after the processing of the front side, because it allows for the new light-transparent substrate to be made of materials that cannot withstand the high temperatures associated with the processing of the front-side of the wafer. Therefore, rather than quartz or sapphire, the new substrate could be a cheaper material such as glass, or for example plastic, which could be much lighter and even flexible.

0245"Thick-Film" SOI and GeOI substrates, share with "Thin-Film" SOI and GeOI substrates the advantages of back-side illumination, such as, wavelength filters made on the back of the buried oxide, and many metal levels with dense interconnects over the pixel areas at the front-side. However, with thicker crystalline silicon or germanium films on top of the buried insulator (typically silicon oxide), there is a stronger absoφtion of the shorter wavelengths in the visible, and UV. Therefore, beyond a certain thickness of the top silicon film, not enough photons of these wavelengths reach the epitaxial layers deposited on that film. For these reasons, it is more advantageous to use Thin-Film SOI and GeOI to fabricate the photo-detector devices described in WO/EPOl/11817, and this will be the substrate of choice for all the exemplary implementations of the innovations described in the present section. Tltin-Film SOI or GeOI CMOS Image Sensors with Back-side Illumination

0246ID or 2D array image-sensors, illuminated only from the back-side of the substrate, present new possibilities for a number of parameters and design options: 1. Wavelength filters are formed on the back of the "buried oxide" of the SOI wafer, once the silicon substrate has been removed, after all processing on the front-side of the wafer has been performed.

02472. Suppression of cross talk between (sub)pixels due to photons impinging on the photo- detectors with wide angles (far from the normal to the substrate). This is a serious problem with conventional image sensors because the color dyes are placed on top of all metal and passivation layers, which might lead to restrictions on the number of metal layers used for image sensor ICs, which in turn could seriously limit the functionality of the design, because certain IP blocks, such as microprocessors or DSPs, require many metal layers.

02483. Possibility of engineering ultra-shallow vertical and/or horizontal pn-junctions for selective UV absoφtion.

02494. Enables PIN &amp; HEP devices with metal (or metal-silicide) top electrodes, forming a Schottky junction on Si and/or SiGeC. hi this configuration, the top metal electrode acts also as a mirror for the HIP device, thereby doubling its absoφtion capability.

02505. Enables the fabrication of layers for the SiGeC/Si Heterojunction Integrated Thermionic (HIT) Coolers on top of the films for the PIN &amp; HIP photo-detectors for "in-pixel" cooling, thus making possible to make olimic and thermal contacts to the HIT devices from the top with metal layers, rather than extracting the heat through a silicon bulk wafer.

02516. Dense interconnects over the area of the photo-detector films, just as if it were simple digital CMOS product. Dense interconnects at the front side, are expected to have a major impact on the overall architecture and design of the entire system, i.e., sensor plus digital logic, memory, etc.

0252Figure 30 shows an example of a Back-Side Illumination configuration for CMOS image sensor made on Thin-Film SOI substrates. The PIN and HIP devices made on the same substrate are monolithically integrated with the CMOS devices.

0253Fabrication Flow

0254The following exemplary process flow is meant to illustrate the most important steps in the simultaneous fabrication of PEN &amp; HIP devices along with the monolithic fabrication of Heterojunction Integrated Thermionic (HIT) coolers, on Thin-Film SOI substrates for back-side illumination.

0255The process steps for the fabrication of PIN &amp; HTP devices together but without the HIT cooler are a subset of the process flow that includes the HIT cooler. The insertion of the process steps pertaining to the present invention take place near the end of what is known as the "Front End of the Line" (FEOL), that is, after all ion-implantation steps have taken place, and prior to the formation of suicide films to make ohmic contacts to the CMOS devices.

0256The electrical isolation between the active areas for CMOS devices and for the photodiodes can be the standard one for the particular CMOS generation that the photodiodes are monolithically integrated with. For such thin top silicon films (e.g. less than 20nm), LOCal Oxidation of Sihcon (LOCOS) process can be used because the "bird's beak" problem cannot develop. After completing the processing for metallization, the removal of the silicon substrate under the buried oxide takes place. The buried oxide of the SOI wafer is used as a marker for the removal process.

0257Before bonding the device layers to a transparent substrate, such as Sapphire, Quartz, Glass or Plastic, several processing options exist: 1. Fabrication of optical cavities for Resonant Cavity Enhanced (RCE) Photo-detection;

02582. Fabrication of color filters for the several wavelengths, with conventional color dye;

02593. Fabrication of Surface Plasmon Polariton (SPP) stmctures.

0260Surface Plasmon Polariton Structures (SPP) stmctures can be used for: 1. Wavelength filtering,

02612. Polarization selectivity,

02623. Coupling of light, especially of long wavelengths such as MWIR &amp; LWIR, to pixels with sub-wavelength lateral dimensions.

0263Process Flow for PIN + HIP + HIT with Back-Side Illumination

0264This Process Flow shows the sequence of steps to fabricate PIN &amp; HIP + HIT, with the PPN &amp; HIP devices having a Schottky top electrode. In this flow it is also shown one possible way to fabricate a contact to the silicon thin-film on top of the buried oxide. The active area where the photodiodes are to be made is separated of surrounding CMOS devices by Field Oxide (FOX) regions. Figure 31 shows a schematic cross-section of a SOI wafer substrate (still with the silicon substrate underneath the buried oxide), after deposition and patterning of a silicon nitiide (Si₃N ) film covering the regions other than the active areas for photo-detectors, hi this figure it shown (thick black line) is a very thin Si0₂ layer under the Si₃N₄ film, whose function is to prevent damage to the silicon surface from the etch process of the Si₃N₄ film.

0265Epitaxial Deposition Module

02661. Epitaxial deposition of the device layers for Schottky PIN &amp; HIP devices and the layers for the HIP cooler. The epitaxial deposition process could be selective or non-selective. In this flow the epitaxial growth will be assumed to be non-selective, thus there will be deposition of amoφhous or poly crystalline layers over the areas covered with a Si₃N film. In the drawing it is shown an epitaxial growth that starts with in-situ doped layers and ends with undoped layers ( Figure 32).

0267Patterning of HIT devices

02682. Photolithography, leaving photoresist only on the regions where the Si/SiGeC layers for the HIT coolers are not to be removed.

02693. Selective etch (wet or dry) of Si/SiGeC layers of HIT cooler, stopping on the buffer layer before the layers of the avalanche region. This buffer layer may contain a marker layer, such as a germanium-rich layer, to provide an etch stop. For example a layer with high concentration of germanium may be incoφorated during the epitaxial growth of the device layers (Figure 33).

0270Patterning of PIN &amp; HEP devices 4. Photo-resist strip and clean.

02715. Photolithography, exposing (removing photoresist) only from the regions where the amoφhous or poly-crystalline Si &amp; SiGeC layers are to be etched away.

02726. Selective etch (wet or dry) of Si &amp; SiGeC material, stopping on the Si₃N₄ film.

02737. Selective etch (wet or dry) of Si₃N₄, stopping onSi0₂ (Figure 34).

0274Spacer Module

02758. Photo-resist strip and clean.

02769. Deposition of Si₃N film with thickness suitable to make spacers for the CMOS devices.

027710. Etch-back of Si₃N₄, stopping on Si0₂ (Figure 35).

0278Suicide Module 11. Removal of thin oxide layer from poly-Si lines and junctions (e.g. with Dilute-HF).

027912. Formation of suicide with conventional methods/recipes (Figure 36).

0280Planarization Module 13. Deposition ofPre-Metal Dielectric (PMD) layer stack.

028114. Planarization of PMD layer stack by Chemical Mechanical Polishing (CMP). (Figure 37).

0282Standard metallization

028315. P atterning of contact holes. 16. Metal filling of contact holes.

028417. CMP of metal (Figure 38).

028518. Full metallization of the front-side of the wafer (not shown in figures). Alignment markers for the patterning on the back-side can be made at a chosen metal level, in selected regions of the wafer.

0286Wafer back-side processing

028719. Attachment of front-side of wafer to a mechanical holder

028820. Removal of silicon substrate. The buried oxide provides a marking layer.

028921. Fabrication of wavelength filters and light-blocking layers on the back-side of the buried oxide (Figure 39).

0290Wafer back-side processing

029122. Attachment of back-side of wafer to a transparent substiate.

029223. Removal of mechanical holder (Figure 40).