Reconfigurable Stochastic neurons based on tin oxide/MoS2 hetero-memristors for simulated annealing and the Boltzmann machine

Abstract

Neuromorphic hardware implementation of Boltzmann Machine using a network of stochastic neurons can allow non-deterministic polynomial-time (NP) hard combinatorial optimization problems to be efficiently solved. Efficient implementation of such Boltzmann Machine with simulated annealing desires the statistical parameters of the stochastic neurons to be dynamically tunable, however, there has been limited research on stochastic semiconductor devices with controllable statistical distributions. Here, we demonstrate a reconfigurable tin oxide (SnOx)/molybdenum disulfide (MoS2) heterogeneous memristive device that can realize tunable stochastic dynamics in its output sampling characteristics. The device can sample exponential-class sigmoidal distributions analogous to the Fermi-Dirac distribution of physical systems with quantitatively defined tunable “temperature” effect. A BM composed of these tunable stochastic neuron devices, which can enable simulated annealing with designed “cooling” strategies, is conducted to solve the MAX-SAT, a representative in NP-hard combinatorial optimization problems. Quantitative insights into the effect of different “cooling” strategies on improving the BM optimization process efficiency are also provided.

Circuit‐Level Memory Technologies and Applications based on 2D Materials

“…For example, the stochastic properties that come from the cycle‐to‐cycle variability in 2D memory cells can be well exploited to build a BM. [ 191 ] The 2D‐material‐based neuromorphic computing systems remain undeveloped compared with that based on oxide‐based memristors, given that fully memristive computing systems have been demonstrated. [ 204 ] The operation of reported 2D‐material‐based systems heavily relies on the incorporation of off‐chip computers or even human intervention.…”

“…e–h) Reproduced under terms of the CC‐BY Creative Commons Attribution 4.0 International license ( https://creativecommons.org/licenses/by/4.0). [ 191 ] Copyright 2021, The Authors, published by Springer Nature.…”

“…Yan et al have made an advancement in utilizing device stochasticity to implement 2D-material-based BMs. [191] Usually, the stochasticity is intentionally suppressed for the sake of small variations and stable circuitry performance. In their work, they exploited the stochastic behaviors of a reconfigurable threeterminal SnO x /MoS 2 hetero-memristor to perform the Fermi-Dirac like sigmoidal sampling (Figure 10e).…”

“…HT refers to an effective high temperature and LT refers to an effective low temperature. e-h) Reproduced under terms of the CC-BY Creative Commons Attribution 4.0 International license (https://creativecommons.org/licenses/by/4.0) [191]. Copyright 2021, The Authors, published by Springer Nature.…”

Memtransistor Devices Based on MoS2Multilayers with Volatile Switching due to Ag Cation Migration

“…[6] RRAMs based on 2D semiconductors as active switching layer have been demonstrated, [7,8] although the advantages of these memory devices over conventional RRAM in terms of scaling or reliability are not clear yet. MoS 2 -based transistors have been reported as selector devices for conventional oxidebased RRAM in one-transistor/one resistor (1T1R) structures, [9,10] although their lowcurrent operation might represent a major limitation over their Si-based counterparts. Charge-based Flash memories have also been developed with 2D semiconductors, [11] although the fundamental issues of Flash in terms of single-electron phenomena and charge retention still represent a concern.…”