QBF-Based Synthesis of Optimal Word-Splitting in Approximate Multi-Level Storage Cells
Daniel E. Holcomb, Kevin Fu · 2014
In applications such as multimedia that tolerate imprecise re-sults, approximate computing techniques can sacrifice pre-cision to save power or time. One aspect of approximate computing is imprecise storage in multi-level cells (MLCs). Computer words that are too large for a single MLC must be distributed across multiple approximate MLCs. The word-level imprecision depends on how the words are split across the MLCs. This work gives an automated synthesis approach for splitting words across MLCs. Given bounds on the im-precision of individual MLCs, the technique synthesizes so-lutions for splitting words across cells to minimize the worst-case imprecision at the word level. The technique is based on quantified Boolean formula solving, within an overall opti-mization loop. Worst-case word-level error is shown to vary by over an order of magnitude across otherwise comparable word-splitting alternatives.