Performance of radiation hardening techniques under voltage and temperature variations

The effectiveness of the techniques to mitigate radiation particle hits in digital CMOS circuits has been mainly studied under a given set of environmental conditions. This paper will explicitly analyze, how the performance of two selected radiation hardening techniques, namely transistor sizing and stack separation, varies with temperature and supply voltage. Our target is an inverter circuit in UMC90 bulk CMOS technology, instances of which have been hardened against charges of 300fC and 450fC using either of the two techniques under investigation. In a Spice simulation we apply particle hits to these circuits through double-exponential current pulses of the respective charge. We study the effect of these pulses in a temperature range from −55 C to +175 C and a supply voltage of 0.65 to 1.2V (nominal 1V) at the output of a (unhardened) buffer that has been connected as a load. For the hardening by sizing we observe proper operation in the range from 1.2V to 900mV, while for lower supply we observe full swing pulses of increasing magnitude when the respective maximum charge is applied. The influence of temperature turns out to be minor. For the stack separation approach the observation is similar, however, the circuit starts glitching only at 750mV. Our study allows the following conclusions: (i) The effectiveness of the hardening approaches strongly depends on the supply voltage, and moderately on temperature. (ii) As expected, low voltage and high temperature represent the worst case for rad-hard sizing. Stack separation, on the other hand, unexpectedly shows a stronger and more complicated temperature dependence. (ii) For voltages below approx. 90% of nominal the hardening by sizing fails, when designed for nominal voltage and room temperature. The approach can be enhanced to survive this worst case by increasing the sizing factor further by more than 3 times. (iv) The stack separation only fails for voltages below approx. 75% of nominal, but there is n- simple remedy to make it reliable for a larger range. This must be considered when judging the appropriateness of this method for a given purpose. Also it turned out that once it fails, the resulting SET pulse has considerable length.