Offset DMR: A Low Overhead Soft Error Detection and Correction Technique for Transform-Based Convolution

A novel concurrent soft error detection and correction scheme is introduced for parallel hardware implementations of transform-based convolution. The proposed technique is based on the structure of radix-2 Fast Fourier Transforms (FFT) of length 2ⁿ where n is an integer. The scheme can provide up to 100 percent detection and correction of isolated single soft errors in the convolution at the cost of little more than double the system area, rather than triple, as is required when using conventional Triple Modular Redundancy (TMR).

1 Introduction

An increasingly important problem in digital circuit design is the detection and correction of soft errors [1]. These transient errors, arising, for example, from single radiation events, can cause stored data bits to be corrupt until new data is written to the memory location. The errors manifest themselves as bit flips at the output of a register or combinational logic gates. Traditionally, soft errors have been a major concern for space applications due to circuit exposure to high radiation levels. With the introduction of smaller device geometries and new process technologies, soft errors are now becoming an issue for ground-level applications as well [1]. Several methods have been proposed to protect circuits from the effects of soft errors. These range from the use of specific manufacturing techniques to the addition of redundancy at the system level to detect and correct the errors [2].