Up

Performance-Driven Functional Decomposition and Circuit Minimization

Abstract

We have developed a circuit minimization method and a functional decomposition method for producing logic circuits with fewer levels. Both are thought to be essential for look-up table (LUT) based field programmable gate array (FPGA) logic synthesis.

Background

When we design a circuit using an LUT based FPGA, we usually first transform the desired functions to an LUT network. The transformed LUT network is thought to be "good" if the number of LUTs and levels are both small. However, since usual logic synthesis methods take account for only basic gates such as AND and OR, they do not always generate good LUT networks. Therefore, we are seeking new methods for generating good LUT networks directly from logic functions.

Performance-Driven Functional Decomposition

An LUT can realize any boolean function with a fixed number of inputs. Therefore, it is important to decompose a large function to small functions for LUT network synthesis. Most of the previously proposed functional decomposition methods are based on the form: f = h(g(X1), X2) where X1 and X2 are disjoint sets of variables. In contrast, we studied a decomposition form: f = h(g1(X1), g2(X2)) (Figure 1). We have developed an efficient method which finds a decomposition form where the total number of variables in X1 and X2 is the smallest. Our method decomposed logic functions with 23% fewer levels than the usual method [1].

Fig. 1: Peformance-driven decomposition
Fig. 1: Peformance-driven decomposition

Circuit Minimization

We have been continuously studying the applications of the circuit minimization method [2], which we developed last year. We plan to use the method in the design flow in Figure 2. The minimization method is applied in this flow to share common sub-functions without increasing levels after the above functional decomposition. Furthermore, the logic expression techniques used in our minimization method were adopted by other researchers who showed their potential.

Fig. 2: Design flow for LUT networks
Fig. 2: Design flow for LUT networks

Future Work

We plan to develop a high-quality logic synthesis method using the above techniques. We also plan to incorporate the developed method into PARTHENON.

References

[1]
Yamashita, S., Sawada, H. and Nagoya, A.: New Methods to Find Optimal Non-Disjoint Bi-Decompositions, Proc. of Asia and South Pacific Design Automation Conference 1998 (ASP-DAC '98), pp. 59-68 (1998).
[2]
Yamashita, S., Sawada, H. and Nagoya, A.: A New Method to Express Functional Permissibilities and Its Application to FPGA Synthesis (in Japanese), IEICE Technical Report, CPSY97-93/VLD97-105, pp. 37-44 (1997).

Contact: Shigeru Yamashita; Email: ger@cslab.kecl.ntt.co.jp


Last modified: Wed Oct 14 15:22:27 1998