Using Error Correcting Codes Without Speed Penalty in Embedded Memories: Algorithm, Implementation and Case Study

Drastic device shrinking, power supply reduction, increasing complexity and increasing operating speeds affect adversely the reliability of nowadays Integrate Circuits (ICs). In many modern designs, embedded memories occupy the largest part of the die and comprise the large majority of transistors. Furthermore, memories are designed as tight as allowed by the process, and are therefore more prone to failures than other circuits. Error correcting codes (ECCs) are an efficient mean for protecting memories against failures. A major drawback of ECCs is the speed penalty induced by the encoding and decoding circuits. In this paper, we present an architecture enabling implementing ECCs without speed penalty. Furthermore, as the manual implementation of this solution is impractical for complex System-on-Chips (SoCs), we propose an algorithm and a set of generic rules allowing automatic insertion of the delay-free ECCs in any complex architecture at Register Transfer Level (RTL). With respect to a naive insertion in the design of the new architecture, the algorithm enable up to 20 % hardware reduction. The Finite State Machines (FSM) that controls the new ECC architecture is also generated automatically. Experimental evaluations show that the hardware overhead of the speed penalty free ECCs protected memory compared to a standard implementation of ECC protected memory is about 2.5 % with an additional power consumption of 6 %.