Mitigation of Single Event Upset (SEU) by Virtual Redundancy in Design by Kaijie Wu*, Polytechnic University, Brooklyn and Jake Karrfalt, Alternative System Concepts, Inc. * presenter Radiation induced Single Event Upsets (SEU) in electronic devices and components, such as encountered in space environments, have traditionally been costly to overcome. One current approach uses triple modular redundancy (TMR), where each processing and control circuit path is tripled and voted, such that when the outputs of the three circuit paths are not the same, assuming single failure, the odd circuit is disregarded. As an SBIR project a low-cost automatic design methodology for achieving radiation hardness was proposed. Through this new algorithmic approach we obtained results similar to TMR. We proved that the algorithmic approach to design can improve radiation tolerance. We developed and tested models of some actual components employing the application of virtual redundancy to error detection, error correction, error diagnosis, and circuit reconfiguration as a means of concurrent error recovery. In Virtual Redundancy the same element is used during idle cycles to re-compute processing functions for concurrent error detection (CED). If the two results do not agree, the computation returns to a checkpoint, and is recalculated and verified. It is thus possible to identify the faulty functional unit. Idle cycles in the design are used for this purpose. Two traditional circuits were selected for analysis and simulation of the results of adding virtual redundancy to the design. The RC6 encryption algorithm was the first example used to validate through simulation the Virtual Redundancy concept. Simulation results, which include injected errors, were effective in proving the concept, yet simple enough to be suitable for detailed analysis. The second example used was to validate the virtual redundancy concept, including response to injected errors, was the FIR filter, another useful design with greater complexity than the RC6 encryption algorithm, providing a more stringent test of idle cycle based CED. The feasibility of developing a tool to create Virtual Redundancy in circuits was proven through simulation during this Phase I program, with a demonstration at SMDC in Huntsville, Alabama on September 12, 2000. The Virtual Redundancy approach is more economical, requires less power, and is more reliable, when compared with traditional TMR, because there is no need to triple the circuit. If a pattern of faults is detected, this circuit can conceivably be used to repair itself. Except for a minor increase in area and power, a radiation hardened circuit can be produced for the same cost, area and power as a non rad-hard commercial component. This has huge payoff in secondary weight effects for power supplies in space applications, but can be used to improve reliability at minimal cost in virtually any application.