Asynchronous techniques for new generation variation-tolerant FPGA

Abstract

This thesis presents a practical scenario for asynchronous logic implementation that would benefit the modern Field-Programmable Gate Arrays (FPGAs) technology in improving reliability. A method based on Asynchronously-Assisted Logic (AAL) blocks is proposed here in order to provide the right degree of variation tolerance, preserve as much of the traditional FPGAs structure as possible, and make use of asynchrony only when necessary or beneficial for functionality. The newly proposed AAL introduces extra underlying hard-blocks that support asynchronous interaction only when needed and at minimum overhead. This has the potential to avoid the obstacles to the progress of asynchronous designs, particularly in terms of area and power overheads. The proposed approach provides a solution that is complementary to existing variation tolerance techniques such as the late-binding technique, but improves the reliability of the system as well as reducing the design’s margin headroom when implemented on programmable logic devices (PLDs) or FPGAs. The proposed method suggests the deployment of configurable AAL blocks to reinforce only the variation-critical paths (VCPs) with the help of variation maps, rather than re-mapping and re-routing. The layout level results for this method's worst case increase in the CLB’s overall size only of 6.3%. The proposed strategy retains the structure of the global interconnect resources that occupy the lion’s share of the modern FPGA’s soft fabric, and yet permits the dual-rail iv completion-detection (DR-CD) protocol without the need to globally double the interconnect resources. Simulation results of global and interconnect voltage variations demonstrate the robustness of the method.