Total-Ionizing-Dose Induced Timing Window Violations in CMOS Microcontrollers

The total-ionizing-dose robustness of low power microcontrollers is investigated. Experiments reveal that with increasing total ionizing dose (TID), the "Timing Window Vio- lations,"i.e., inability of the instruction set to execute within the clock-cycle(s) lead to failures in microcontroller operations. Clock frequency and supply voltage of the microcontroller are varied to determine the maximum clock frequency at which the micro- controller can execute software subroutines without failure. Low power microcontrollers from two different manufacturers were tested. The maximum clock frequency decreases with increasing TID for both parts. A model for the degradation based on analysis of circuit level timing models is presented. The microcontroller robustness implications for system designers and ASIC designers are discussed. TID has been shown to impact propagation delays in indi- vidual transistors (4), the timing characteristics of integrated circuits (5), memory access times (6), and propagation delays in CMOS flash-based FPGAs (7). In this work, timing window violations are experimentally demonstrated to be the primary source of failure in response to TID for a class of low power microcontrollers, and a model for the degradation and hard- ening implications are presented. The conclusion that timing window violations are the primary source of failure is supported by a measurement technique that allows insight into the in- ternal degradation of the device during the radiation exposure. The technique is necessary when information regarding indi- vidualtransistordegradation,themicrocontrollertechnology,or the fabrication process is not available, such as incorporating a COTS mechatronic subassembly including an integrated mi- crocontroller. Timing window violations are identified by per- forming a series of software tests on the microcontroller at var- ious frequencies and supply voltages. The technique to deter- mine the maximum and minimum frequencies and voltages for the microcontroller is more commonly used in electrical char- acterization (8). Results are presented for a commercially avail- able microcontroller, the Atmel ATMEGA328P (9). Analysis of the experimental results shows that the degradation is con- sistent with the expected degradation of logic gate switching time based on charge build-up in the gate oxide and shallow trench isolation (10). The results are further verified using a subset of the software tests on a Microchip PIC16F677 (11). The clock frequency measurement is nondestructive and can be monitored in a field deployment scenario to evaluate the health of the device, allowing for mission planning as well as life extensionthrough reducingclockoperating frequencyor in- creasing supply voltage.