Approach to the design of parity-checked arithmetic circuits

Achieving fault tolerance via parity checking is attractive due to low overhead in storage and interconnect. However, nonpreservation of parity during arithmetic operations makes it necessary to strip the parity bit before, and to restore it after, such operations. This either leaves the arithmetic part unprotected or else requires complex code conversions. We show that some redundant representations, which are often used for high performance anyway, support a way of designing low-overhead, fault-tolerant arithmetic hardware circuits. An added benefit is localized fault effects due to carry-free arithmetic. Our proposed fault tolerance strategy consists of a way of converting parity-encoded input values to even-parity redundant representations, performing arithmetic with redundant operands in such a way that parity is preserved, and, finally, converting any redundant result to standard parity-encoded output.