hamed wrote on Jun 14th, 2006, 6:23am:Hi all,
I think one has to be careful when defining "worst case" for analog or digital designs.
While in analog circuits, Vth is the most meaningful parameter which "Slow Cornet" is defined based on that, in digital circuits gds and capacitors determine performance of the circuit in terms of the speed, and Vth is almost meaningless. That's why the "slow corner" gets worse if temperature decreases (Vth increases).
In addition there are some cases in which the interaction of devices behaviour changes the whole performance in a direction that may looks strange.
Hamed
Hi Hamed,
I would be wary of making a statement like that. The most important parameter in analog design is arguably the transconductance (Gm) of the device. In fact, even in digital circuits, which are analog in reality, the same holds equally true.
And Gm depends first on mobility of the device (Kn'= un*Cox). Vth is a composite of several quantities, and ideally ought not to influence the speed. However, a larger Vth implies a higher overdrive to achieve the same Gm in the device. So both un and Vth are instrumental here.
The fundamental speed of the device is proportional to un/L, and hence shorter channels offer higher speed due to reduced transit time.
Please keep in mind the following:
1. Vth variation across process is strongly dependent on doping and oxide contamination as far as I know (Vth here implies threshold voltage).
2. If wafer A and B have the same un (mobility) & Cox & L but differ only in Vth, then the wafer with the larger Vth is the slower one. However, Vth alone cannot provide speed information.
3. Given a wafer, the speed will typically reduce with temperature despite the reduction in Vth at higher temperature. This is due to the fact that the mobility reduction due to higher temperatures has a stronger impact on speed than Vth reduction. Hence, the worst-speed situation is for worst-speed process (highest Vth, lowest Kn') and highest temperature with lowest supply.
4. Thus, the worst-speed corner assumes lowest un, highest Vth, and also longest L (in good process models).
5. Perhaps there are some special circuits which behave a bit differently, but in general the device transit time should increase as the temperature goes up, and this is the fundamental criteria for testing the speed of a device.
Regards
Vivek