Integrated circuits (ICs) are increasingly susceptible to uncertainty caused by soft errors, inherently probabilistic devices, and manufacturing variability. As device technologies scale, these effects become detrimental to circuit reliability. In order to address this, we develop methods for analyzing, designing, and testing circuits subject to probabilistic effects. Our main contributions are: (1) a fast, soft-error rate (SER) analyzer that uses functional-simulation signatures to capture error effects, (2) novel design techniques that improve reliability using little area and performance overhead, (3) a matrix-based reliability-analysis framework that captures many types of probabilistic faults, and (4) test-generation/compaction methods aimed at probabilistic faults in logic circuits.
SER analysis must account for the three main error-masking mechanisms in ICs: logic, timing, and electrical masking. We relate logic masking to node testability of the circuit and utilize functional-simulation signatures, i.e., partial truth tables, to efficiently compute estability (signal probability and observability). To account for timing masking, we compute error-latching windows (ELWs) from timing analysis information. Electrical masking is incorporated into our estimates through derating factors for gate error probabilities. The SER of a circuit is computed by combining the effects of all three masking mechanisms within our SER analyzer called AnSER.
Using AnSER, we develop several low-overhead techniques that increase reliability, including: (1) an SER-aware design method that uses redundancy already present within the circuit, (2) a technique that resynthesizes small windows of logic to improve area and reliability, and (3) a post-placement gate-relocation technique that increases timing masking by decreasing ELWs.
We develop the probabilistic transfer matrix (PTM) modeling framework to analyze effects beyond soft errors. PTMs are compressed into algebraic decision diagrams (ADDS) to improve computational efficiency. Several ADD algorithms are developed to extract reliability and error susceptibility information from PTMs representing circuits.
We propose new algorithms for circuit testing under probabilistic faults, which require a reformulation of existing test techniques. For instance, a test vector may need to be repeated many times to detect a fault. Also, different vectors detect the same fault with different probabilities. We develop test generation methods that account for these differences, and integer linear programming (ILP) formulations to optimize test sets.