As the feature size of LSI becomes smaller, random variation of each element becomes more influential. The variation of elements decreases the yield, and makes conventional design based on worst-case estimation too pessimistic. The mechanism of detecting and recovering timing-faults makes the most of the advantages of becoming smaller feature size of LSI and make high-performance and low power operation possible.
In this study, we have proposed a technique that enables dynamic time borrowing which cuts the overhead of recovering error by tolerating accumulation of timing-delay. Our technology realizes more efficient timing-fault tolerant VLSI than conventional related technologies. We can realize this technology by combining two-phase latch system and dynamic timing-fault detection. We verify the timing-fault tolerant processor which is implemented with introducing this technique into critical path, and make a study of automation technique of the latch insertion to a critical path.
We are now developing a program for applying the proposed method to existing circuits. We try to implement and evaluate the search algorithm for determining the position of the latch insertion.