| Budget Amount *help |
¥3,700,000 (Direct Cost: ¥3,700,000)
Fiscal Year 2002: ¥600,000 (Direct Cost: ¥600,000)
Fiscal Year 2001: ¥600,000 (Direct Cost: ¥600,000)
Fiscal Year 2000: ¥2,500,000 (Direct Cost: ¥2,500,000)
|
|---|
| Research Abstract |
Using dual-rail coding for data-path circuits is one of the major approaches for asynchronous circuit design. This approach requires a lot of C elements, which are memory devices, in control circuits, and it causes several problems in synthesis and verification, such as increase of state spaces, degradation of performance, and so on. Thus, this project proposes using three-rail coding instead of dual-rail coding to implement circuits without C elements. Actually, we have selected a ternary code for such the three-rail coding, and have developed two procedures to obtain basic ternary gates from given truth tables based on the ternary code. Furthermore, some optimization technique has also been developed.
On the other hand, when we want to verify an asynchronous circuit including data-paths, we usually do the abstraction of the circuit to remove most of those data-paths. This is because verifying the circuit without abstraction requires checking the whole data values that the data-paths take, and this makes it too difficult to describe specifications and traverse its huge state space. This project also proposes to verify data-paths partially using some specific or random values with verifying the control parts formally as usual. This allows us to do the verification of a given circuit without abstraction which still gives us rather reliable results. For this purpose, we have extended the existing model (a time Petri net) such that it can handle producing and comparing data. In addition, in order to avoid the state explosion problem, a partial order reduction algorithm, which can efficiently prune away the state spaces unnecessary for the verification and synthesis, for such an extended time Petri net has been developed. According to several benchmark circuits, it has been found to outperform a verifier based on the existing model.
|
