MTBF Estimation in Coherent Clock Domains

Abstract

Abstract-Special synchronizers exist for special clock relations such as mesochronous, multi-synchronous and ratiochronous clocks, while variants of N-flip-flop synchronizers are employed when the communicating clocks are asynchronous. N-flip-flop synchronizers are also used in all special cases, at the cost of longer latency than when using specialized synchronizers. The reliability of N-flip-flop synchronizers is expressed by the standard MTBF formula. This paper describes cases of coherent clocks that suffer of a higher failure rate than predicted by the MTBF formula; that formula assumes uniform distribution of data edges across the sampling clock cycle, but coherent clocking leads to drastically different situations. Coherent clocks are defined as derived from a common source, and phase distributions are discussed. The effect of jitter is analyzed, and a new MTBF expression is developed. An optimal condition for maximizing MTBF and a circuit that can adaptively achieve that optimum are described. We show a case study of metastability failure in a real 40nm circuit and describe guidelines used to increase its MTBF based on the rules derived in the paper.

Measurement of Setup and Hold Time in a CMOS DFF for a Synchronizer

“…where   and   are the clock and data transition frequencies respectively, s is the time allowed for metastability resolution,  is the resolution time constant, and   is the parameter describing a vulnerable time window [5][6].…”