Abstract
This article analyzes the effect of imperfections in physically realizable memory. Motivated by the realization of a bit as a Brownian particle within a double well potential, we investigate the energetics of an erasure protocol under a Gaussian mixture model. We obtain sharp quantitative entropy bounds that not only give rigorous justification for heuristics utilized in prior works, but also provide a guide toward the minimal scale at which an erasure protocol can be performed. We also compare the results obtained with the mean escape times from double wells to ensure reliability of the memory. The article quantifies the effect of overlap of two Gaussians on the the loss of interpretability of the state of a one bit memory, the required heat dissipated in partially successful erasures and reliability of information stored in a memory bit.
Highlights
Over the last four decades, the semiconductor industry has made significant headway in improving the performance of complementary metal oxide semiconductor field effect transistor (CMOS-FET) devices, while consistently reducing their size
We quantify that a trade-off exists between lowering the minimum heat dissipation in information erasure and improving reliability of information stored in a single bit memory
The Generalized Landauer Bound (GLB) from [7,14,15] states that, if the reliability parameter of the erasure process is p, the associated average heat dissipation is at least k B T (ln 2 + p ln p + (1 − p) ln(1 − p))
Summary
Over the last four decades, the semiconductor industry has made significant headway in improving the performance of complementary metal oxide semiconductor field effect transistor (CMOS-FET) devices, while consistently reducing their size. We analyze the effect of overlap of the two wells in a single bit memory with regards to heat dissipation in erasure of information as well as reliability of stored information. We obtain complimentary lower bounds on the decrease in thermodynamic entropy, demonstrating that these bounds are reasonably sharp, and, for bi-stable wells, physically separated by lengths close to their standard deviation, the error in entropy approximation incurred by the “insignificant overlap” approximation is significant These quantitative results are of immediate application as they allow a tight approximation of the change in entropy in erasure process, relevant to the precise estimation of the associated minimal heat dissipation. We quantify that a trade-off exists between lowering the minimum heat dissipation in information erasure and improving reliability of information stored in a single bit memory. The analysis is motivated by new paradigms of computation—for example, in stochastic computation [20] and neuromorphic memory architectures [21], where uncertainty of the success of computation is allowed and modeled a priori
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
