Soft Error Rates with Inertial and Logical Masking

Abstract

We analyze the neutron induced soft error rate (SER). An induced error pulse is modeled by two parameters, probability of occurrence and probability density function of the pulse width. We calculate failures in time (FIT) rates for ISCAS85 benchmark circuits. A comparison with measured SER for SRAMs shows better relevance of our work over other published work. Our CPU times are reasonable; benchmark circuit C1908 with 880 gates requires only 1.14seconds. Further, we study the influence of circuit topology on SER. We find that for some circuits with many levels of logic there exists a critical single event transient (SET) width. For smaller induced pulse width the SER depends not on the size of the circuit but only on the gates near the output, and only those need to be protected. For an inverter chain in TMSC035 technology, the critical width is between 25ps and 50ps. For a shallow circuit, e.g., a ripple-carry adder, the critical SET width may not exist.