NMOS Subthreshold RVT

The same test bench was then used to perform the same geometric sweep with the supply voltage lowered to 250mV (VT = 402mV). Test cases were performed for SS/TT/FF process corners. The test bench was then modified to connect the gate terminal to ground and the same sweeps performed to determine the leakage current. Finally the data of both test benches was used to calculate the Ion/Ioff ratio for the sweeps. This data can be seen in Figures 5, 6 and 7.

Figure 5 shows the vestiges of the short channel effect still manifest in the drain currents across all process corners. In a self-aligned deep submicron bulk planar process, high dopant density source and drain regions are implanted via an ion implantation step, as described in the previous chapter. Due to the [100] orientation of the silicon

monocrystalline lattice structure (used to minimize threshold variation by reducing variations in surface state potential), the wafer is angled at around 7% during ion implantation to prevent channeling (dopants travelling deep into the substrate due to collision avoidance with the lattice structure). Ion implantation is also used to create the retrograde dopant profile of the channel in order to prevent bulk latch up and set up the characteristics of the device. In terms of global variation, the variation in channel dopant depth and overall dopant density in the channel from the above processing step is one of the differences between the SS/TT/FF process corners.

In the SS corner where the channel dopant has remained relatively close to the surface of the device, the dopant density at the channel surface is greater. This requires more voltage at the device’s gate to invert it. At minimum width, the SS sweep in Figure 5 shows a typical RSCE, with the drain current increase in response to an increase in length, contrary to superthreshold operation. The higher dopant density means that the INWE effect becomes less observable, but is still present in the width response, defying the linear relationship by offering a greater drain current per unit width as the width tends towards the minimum. Bends and quirks are apparent in the SS and FF plots where smooth continuation would be expected. This is simply a result of the approximation of the actual physics by compact equations in the compact BSIM models. The apparent observable discontinuation is simply a result of the model transitioning from one equation to another.

In the TT sweep where the dopant has successfully created the intended channel profile, a balance between the opposing SCE and RSCE physical effects is observed at minimum width. Deviation in the linearity of the width response is more observable, with a larger increase in drain current per unit width as the width tends towards the minimum.

In the FF sweep where the dopant has channeled deeper into the substrate, the dopant density at the surface of the device is less, requiring less voltage at the gate to invert the channel. The SCE dominates due to the diminishing influence of the HALO dopants that create the RSCE. The monotonic rise in drain current as the length tends towards

minimum length therefore resumes. However, the lower dopant density at the surface of the channel means that the additional fillip provided by the INWE is greater, resulting in

not only a deviation in the linear current-width proportionality, but an increase in drain current as the device width tends towards minimum.

Figure 6 shows the leakage current sweeps across the three process corners. Due to the underlying physics, similar features are observed in the leakage current characteristics as are observed in the drive current characteristics. In the SS corner, the HALO implant s have less impact on the channel current when the device is off. The balance between the SCE and RSCE is therefore favored more towards SCE. More impact is observed from the INWE with the deviation in the linear width-leakage current relationship deviating greater than in the drive current sweep. In the TT sweep, the SCE has already begun to dominate the length response, with little sign of the RSCE. The INWE is more prominent than in the drive current sweep. Finally in the FF sweep, SCE is completely dominant with a clearly observable monotonic rise in leakage current as the length tends towards the minimum. The INWE is also prominent, with a substantial rise in leakage current as device width tends towards minimum.

By comparing the results from both figures, several overall trends are observed. Firstly, the RSCE is most prominent in the slow (SS) corner and least prominent in the fast (FF) corner. This is logical, given that the underlying physical effect is caused by increased dopant density via HALO implants (Section 2.5.2.3). Should these additional dopants channel deeper into the substrate or have fewer dopants (both occur as process tends to FF) then the impact of the RSCE will begin to diminish. Secondly, the INWE is most prominent in the fast corner and least prominent in the slow corner. This again is logical, given that the underlying physical effect is a fillip analogous to a thinning of the gate oxide at the orthogonal sidewalls (Section 2.5.7). If the dopant density at the surface of the channel is lowered, a deeper depletion and inversion is induced from the additional electric field strength dropped across the gate oxide translating into an increase in current. The typical (TT) corner is therefore a balance between these two physical effects.

Thirdly, the impact of the effects are slightly different in the on and off currents. This therefore warranted further processing in the form of Ion/Ioff ratio sweeps, an important metric when determining performance/leakage or performance/energy.

Figure 7 shows similar trends in the Ion/Ioff ratio across all three-process corners. The SS corner sweep shows that Ion/Ioff ratio rolls off as the device length approaches minimum.

There is also a distinct increase in the Ion/Ioff ratio as the width approaches minimum width, after the peak at around 230nm in length (Where the RSCE diminishes). The TT corner sweep shows that the Ion/Ioff ratio falls off as the device length approaches minimum, however this is deeper than in the SS corner. No increase is observed at the width approaches minimum as was observed in the SS corner.

The FF corner sweep shows the same fall off in Ion/Ioff ratio as the length approaches the device minimum but is even deeper still. This time a decrease in Ion/Ioff ratio is observed as the width approaches minimum after the RSCE peak at around 180nm.

These trends suggest several relationships. The first is that the SCE has a greater impact on leakage current than on the drive current. Therefore this impoverishes the Ion/Ioff ratio as the length tends towards minimum. Given that the SCE is least prominent in the slow corner, the fall off in the Ion/Ioff ratio is least at this corner and greatest in the fast corner.

The second relationship observed is that the INWE also has a greater impact on leakage current than on the drive current. This is shown by the increase to decrease trend in the Ion/Ioff ratio at minimum width as the process corners tend from SS to FF.