Adaptive threshold read algorithms in multi-level non-volatile memories with uncertainty

Abstract

For an array of non-volatile memory cells read by threshold measurements, read performance is determined by the number of measurements needed to read the memory word. In this work we wish to explore the time savings when the measurement sequence is allowed to terminate with some prescribed amount of uncertainty. This residual uncertainty will be treated by a specially designed code. The model we introduce in this work allows the read process to terminate the measurement sequence when up to T cells have their levels known only up to a set of W possible levels. For this model we propose efficient read algorithms, we analyze the expected number of read measurements, and plot results for interesting parameters. Our results show that allowing uncertainty in the readout and cleaning this uncertainty with a special code gives a better read performance compared to a non-redundant representation that is read at full without uncertainty.