[Back]

M Függer, A. Dielacher, U. Schmid:
"How to Speed-Up Fault-Tolerant Clock Generation in VLSI Systems-on-Chip via Pipelining";
Talk: EDCC - 8 (European Dependable Computing Conference), Valencia, Spain; 2010-04-28 - 2010-04-30; in: "Proceedings of the Eight European Dependable Computing Conference", IEEE Computer Society, (2010), ISBN: 9780769540078; 230 - 239.

Fault-tolerant clocking schemes become inevitable when it comes to highly-reliable chip designs. Because of the additional hardware overhead, existing solutions are considerably slower than their non-reliable counterparts. In this paper, we demonstrate that pipelining is a viable approach to speed up the distributed fault-tolerant DARTS clock generation approach introduced in (Függer, Schmid, Fuchs, Kempf, EDCC'06), where a distributed Byzantine fault-tolerant tick generation algorithm has been used to replace the traditional quartz oscillator and highly balanced clock tree in VLSI Systems-on-Chip (SoCs). We provide a pipelined version of the original DARTS algorithm, termed pDARTS, together with a novel modeling and analysis framework for hardware-implemented asynchronous fault-tolerant distributed algorithms, which is employed for rigorously analyzing its correctness & performance. Our results, which have also been confirmed by the experimental evaluation of an FPGA prototype implementation, reveal that pipelining indeed allows to entirely remove the adverse effect of large interconnect delays on the achievable clock frequency, and demonstrate again that methods and results from distributed algorithms research can successfully be applied in the VLSI context.

http://dx.doi.org/10.1109/EDCC.2010.35