Method of manufacturing an integrated circuit having p-MOSFETs with different channel widths

TECHNICAL FIELD

The field of the invention in that of buried channel p-FET integrated circuit processing in which the transistors are isolated by trenches and in which the groundrules are in the range of about 0.25µm.

BACKGROUND OF THE INVENTION.

It has been known that corner conduction in trench isolated n-MOSFETs may be a significant contributor to standby current in low standby power ULSI applications, as described in A. Bryant, et al "The current-carrying corner inherent to trench isolation", IEEE Electron Device Letters, vol 14, no. 8, pp. 412-414, 1993 and B. Davari, et al "A variable-size shallow trench isolation (STI) technology with diffused sidewall doping for submicron CMOS", 1988 IEDM Technical Digest, pp. 92-95, 1988. A manifestation of corner conduction is inverse narrow channel effect when the standard current definition of V_t (I_Vt = 40x(W_des/L_des)nA) is applied. However, corner conduction in buried-channel p-MOSFETs has not, heretofore, been recognized as a concern. In buried-channel p-MOSFETs, the polarity of the work function difference between the N+ poly gate and the buried p-layer depletes the buried layer of carriers at low gate voltages. Due to the geometrically enhanced field at the silicon corner, it is expected that, when the doping is uniform across the device width, the magnitude of V_t at the corners of these devices is greater than at mid-channel. This leads to a normal narrow channel effect.

SUMMARY OF THE INVENTION

The invention relates to a CMOS integrated circuit in which a performance advantage for p-FETS is achieved by use of an anomalous narrow channel effect.

A feature of the invention is the use of wide transistors having a V_t chosen to meet off-current design objectives together with narrow transistors having a reduced V_t for improved performance.

Another feature of the invention is the use of improved performance high-current transistors comprising a number of narrow transistors connected in parallel.

BRIEF DESCRIPTION OF THE DRAWING

Figure 1 illustrates measured and predicted values of V_t for different buried-channel transistor widths.

Figure 2 illustrates the dependence of V_t on channel width for buried-channel p-FET having channel widths in the range of 2 -20µm.

Figure 3 illustrates the distribution of Boron concentration in a buried-channel p-FET.

Figure 4 illustrates a high-current low V_t transistor set.

BEST MODE OF CARRYING OUT THE INVENTION

An anomalous narrow channel behavior has been observed on buried-channel p-MOSFETs based on a 0.25 µm CMOS process. The magnitude of the threshold voltage decreases when going from very wide devices to approximately 0.4 µm width (referred to as the inverse narrow channel effect). For devices narrower than 0.4 µm, the expected normal narrow channel effect is observed' as described in K. Ohe, et al "Narrow-width effects of shallow trench-isolated CMOS with n⁺ polysilicon gate!, IEEE Trans. Electron Devices, vol. 36, no. 6, pp. 1110-1116, 1989. Fig 1 shows the relationship between V_t and channel width for buried-channel p-FETs having gate oxide thickness of about 10nm and V_DS of 0.1V. These characteristics are also typical of transistors of shorter channel length and higher drain-source voltage.

Fig. 2 illustrates the relationship for a wider range of channel width. If the buried-channel boron dose is decreased to assure that the off current is less than or equal to the design objective off current over the entire range of design widths, as would be consistent with the prior art, the performance of the widest devices is compromised as a result of higher than desired V_t. A loss of overdrive of 100mv on the widest devices is typical.

A performance advantage based on the anomalous narrow channel effect can be realized by using relatively narrow (2 -10µm) devices for the logic circuitry and wide (> 20µm) devices for the inverters that drive large loads by setting the V_t of the widest devices to meet the off-current objective, which permits a reduction in the magnitude of V_t for the narrower devices, thus giving improved performance. As is known, the V_t for the devices is initially set by the buried channel implant, conventionally done after the deep (or latchup) implant. The deep implant then raises V_t for the wider devices. Since most of the standby current is contributed by the wide devices, this approach has little effect on total standby current.

It has been found that the anomalous narrow channel effect results from a "boron puddle" or localized increase in the boron concentration near the edge of the isolation trenches created by transient enhanced diffusion (TED) of the boron in the buried layer. The TED, in turn, results from an excess interstitial (described in "Paint Defects and Diffusion in Silicon", P. M. Fahey, et al Reviews of Modern Physics, vol 61, no. 2, April 1989) distribution arising from damage caused by the 2.5x1013/cm2, 500 keV phosphorous implant used for latchup suppression, in conjunction with strong excess interstitial recombination on the top surface and trench sidewalls of the device.

The performance improvement due to the anomalous narrow channel effect has been modelled using SPICE and it has been found that, for a typical delta V_t of 100mV between devices having p-channel width W_p of Wp = 2µm and Wp = 20µm, there is a performance improvement of 4.7% using NAND logic. If the delta V_t is increased to 200mV by enhancing the anomalous narrow channel effect, the performance improvement increases to 9.3%. Even greater performance improvements can be obtained using NOR logic, without any increase in standby current.

The anomalous narrow channel effect can be enhanced by increasing the implanted point defect concentration. These point defects provide the source of the excess interstitials that drive the formation of the boron puddle and therefore the anomalous narrow channel effect. One method of enhancement is that of increasing the dose of the deep phosphorous implant. Such an increase does not have an effect on device operation because the dopant provided by this implant is too deep ( ∼ 0.8µm) to affect the device. Indeed, the increase in dopant also improves latchup suppression. Doubling the deep implant dose at the same energy is shown in simulations to increase the delta V_t (between Wp = 2 to 20µm) from 100mV to 200mV.

For purposes of the claims, the term "latchup dose" will be taken to mean an implant of 2.5x10¹³/cm2 at an energy of 500keV and the term "latchup-threshold dose" will be taken to mean a dose greater than 4x10¹³/cm at an energy of greater than 500keV. Also, the terms "narrow" and "wide" when applied to the p-FET channel width will be taken to mean less than about 10µm and greater than about 10µm, respectively.

In addition, the energy of the deep implant can be increased (illustratively to 1MeV) for even greater enhancement by increasing the crystal damage caused by the implant. Additionally, the use of higher implant energies moves the tail far enough from the surface that even greater doses can be used without affecting the device doping. Implanting other species such as silicon or any other species that is not electrically active can also be used to increase the anomalous narrow channel effect without affecting the latchup effect of the deep implant.

If, on the other hand, it is desired to reduce the anomalous narrow channel effect, the addition of an annealing step between the deep implantation and the gate oxidation of 60 minutes at 800°C provides for rapid excess interstitial recombination on the Si/SiO2 interfaces, which greatly reduces the TED of boron during the 850°C gate oxidation. Alternatively, the gate oxide (illustratively about 10nm) may be formed at a higher temperature such as 1050°C for a correspondingly shorter time that reduces the TED.

In a conventional process, a 15 minute, 850°C dry gate oxidation immediately follows the n-well and boron buried-channel implants. An 800°C 60 minute anneal immediately following the deep phosphorus implant and prior to the gate oxidation results in rapid excess interstitial recombination on the Si/SiO2 interfaces. Thus, TED of boron is greatly reduced during the subsequent 850°C gate oxidation. Measurements and simulation confirm that the "boron puddle" and anomalous narrow channel effect are greatly reduced (net p-type concentration peaks only 3% above mid-width value for a device width of 2µm). The boron diffusivity at the depth of maximum boron concentration (∼ 30nm) increases and the boron distribution broadens from trench edge toward the middle of the area between the trenches ( the active area). The minimum in the peak net p-type concentration of the buried layer occurs at mid-width and is accentuated by the compensation of boron by the n-well doping. Furthermore, the gradient in the excess interstitial concentration drives the boron towards the trench edge. Within approximately 30 nm of the trench edge, segregation induced boron depletion and n-type dopant pile-up reduce the net p-type concentration.

The overall effect of the excess interstitial distribution on the diffusion of boron during the 850°C gate oxidation is the creation of a "boron puddle" near the trench edge, as shown by the simulated net doping distribution at end of process in Figure 3 for a channel width of about 0.9µm.

Modeling shows that the peak net p-type concentration in the "boron puddle" is approximately 32% greater than at mid-width, for a 2 µm wide device. As the device width is reduced down to approximately 0.4 µm, the peak net p-type concentration increases because of decreased interstitial concentration at the top surface resulting from recombination on the trench sidewall. For still narrower devices the effect of the n-type dopant pileup on the trench sidewall reduces the buried layer concentration. Figure 3 shows measured V_t vs W characteristics compared with simulated results obtained from TSUPREM-4 and FIELDAY. It is noteworthy that evidence of a comparable boron puddle for the complementary n-MOSFET is neither observed experimentally, nor predicted by modeling. The use of the extra anneal results in a drastically reduced "boron puddle" and also results in a lower magnitude of wide device V_t, since reduced TED of boron leaves a higher peak net p-type concentration in the buried layer.

In addition to setting the deep implant parameters to provide low leakage for wide devices and enhance performance for logic devices, it is also possible to take advantage of the dependence of V_t on width to provide for low V_t where it is needed for high-current applications by connecting a number of narrow channel devices in parallel to provide the desired total current. Fig 4 illustrates in partially schematic, partially pictorial form a set of transistors denoted generally with the numeral 100 having a source node 30, a drain node 40 and a common gate 20. Each of the N transistors in the set 10-1 to 10-N has a channel width Wn chosen to provide the desired V_t. The total number N is chosen to provide the required current for the application in question.

Those skilled in the art are well aware that there are many steps in an integrated circuit process. For convenience in terminology, the phrase "preparing the substrate" will be used to indicate conventional steps such as growing an epitaxial layer (where used), performing a threshold implant, cutting and filling trenches for isolation, forming p-and n- wells etc. Similarly, the phrase "completing the circuit" will be used to indicate the remaining steps after the deep implant such as completing the transistors p-FET and/or n-FETs and forming the interconnections, etc.