Panel: Deep Submicron Design Challenges

摘要

This panel will provide access to a group of ASIC engineers familiar with the problems of deep submicron design that will face us in the near future. As we move to deep submicron CMOS processes we are scaling the metal linewidth but not the thickness. This is increasing the parasitic capacitance of the wires to the point where the typical interconnect delay is larger than a gate delay. At the moment synthesis tools work by minimizing the number of literals in Boolean logic expressions and mapping to gates in a cell library with the smallest databook delay values. Instead of concentrating on logic synthesis and pre-layout simulation and adding the wires, almost as an afterthought, we will need to think far more carefully about interconnect. One of the ways to do this is to couple the synthesis and physical layout steps far more tightly than they are now. It is not at all clear how best to do this. Interconnect is one of the biggest, most visible, and perhaps most easily understood problems in deep submicron design, but there are others. In fact, there are many second- and third-order physical effects that are starting to become more important. In the past we have swept these problems under the carpet by using worst-case design. We can no longer afford the increasing number and increasing size of these pessimistic assumptions. Some other examples of effects we will need to consider: We will have to pay more attention to the shape of logic waveforms and not just their delay. Many ASIC vendors have already started to account for the effect of rise and fall times. We will need to develop methods to tackle interconnect coupling. We cannot afford to extract all coupling capacitances. With deep submicron processes we are moving into an era where we need low-power design, since batteries and packages donÕt seem to be keeping up. We need new tools and methods to help us in this area. Lithography is becoming very difficult as we run out of room in the light spectrum. At -the moment we put what we want on a mask and expect the lithography engineers to reproduce it on silicon. This may not continue to be possible.