Reversible computing with mechanical links and pivots

With the concern that “Moore’s Law is dead,” new interest has grown for unconventional forms of computing. This includes: quantum computing (for atomic simulations and specific NP approximation algorithms) analog computing (for fast-and-efficient-but-error-prone computation, such as neural net inference) reversible computing (for maximum energy efficiency) It is believed that the most efficient computing devices would use little or no entropy during reversible computations, and only consume energy during non-reversible parts of computation. Specifically, the Landauer’s principle states that all non-physically-reversible computation operations consume at least \(2.9 \times 10^{21}\) J of energy at room temperature (and less as the temperature drops). Are we at the theoretical maximum efficiency yet? No. Using some back-of-the-napkin math: AMD Ryzen Threadripper 7995WX has about 12 TFLOPS of FP32 (single-precision) compute (link) Assuming 30 irreversible computations per FP32 oper ... Read full article.