An atomic Boltzmann machine capable of self-adaption

Abstract

The Boltzmann Machine (BM) is a neural network composed of stochastically firing neurons that can learn complex probability distributions by adapting the synaptic interactions between the neurons 1 . BMs represent a very generic class of stochastic neural networks that can be used for data clustering, generative modelling and deep learning 2 . A key drawback of software-based stochastic neural networks is the required Monte Carlo sampling, which scales intractably with the number of neurons. Here, we realize a physical implementation of a BM directly in the stochastic spin dynamics of a gated ensemble of coupled cobalt atoms on the surface of semiconducting black phosphorus. Implementing the concept of orbital memory utilizing scanning tunnelling microscopy, we demonstrate the bottom-up construction of atomic ensembles whose stochastic current noise is defined by a reconfigurable multi-well energy landscape. Exploiting the anisotropic behaviour of black phosphorus, we build ensembles of atoms with two well-separated intrinsic time scales that represent neurons and synapses. By characterizing the conditional steady-state distribution of the neurons for given synaptic configurations, we illustrate that an ensemble can represent many distinct probability distributions. By probing the intrinsic synaptic dynamics, we reveal an autonomous reorganization of the synapses in response to external electrical stimuli. This self-adaptive architecture paves the way for on-chip learning directly in atomic-scale machine learning hardware.

Ising machines as hardware solvers of combinatorial optimization problems

“…(e) Co atoms on surface of black phosphorus interacting with a scanning tunneling microscope to realize a Boltzmann machine from Ref. 33 . (f) Spatial light modulator based photonic annealer from Ref.…”

“…When implemented on dedicated hardware, this provides the chance to exploit the parallelization of digital hardware accelerators and analog computing. For the analog computation approach, numerous physical implementations of Ising and related models have been realized or proposed, including magnetic devices 29,[44][45][46][47][48][49][50][51] , optics 34,52,53 , memristors 30,54 , spinswitches 55 , quantum dots 56 , single atoms 33 , microdroplets 57 , and Bose-Einstein condensates 24,58 (see Fig. 2).…”

Convolutional neural networks with radio-frequency spintronic nano-devices

“…Spintronics, whose memory, multifunctionality and dynamics are attractive for information processing is another technology actively studied for neuromorphic computing. Spintronic devices present advantages for the integration since they are compatible with CMOS and the same technology can be used to implement both artificial neurons and synapses [22,23]. Research shows that arrays of spintronic memories are promising for associative memories [24], spiking neural networks [25] and convolutional neural networks with time-domain computing [26].…”

Precise atom manipulation through deep reinforcement learning

“…

Atomic-scale manipulation in scanning tunneling microscopy 1 has enabled the creation of quantum states of matter based on artificial structures 2-4 and extreme miniaturization of computational circuitry based on individual atoms. [5][6][7] The ability to autonomously arrange atomic structures with precision will enable the scaling up of nanoscale fabrication 8 and expand the range of artificial structures hosting exotic quantum states. However, the a priori unknown manipulation parameters, the possibility of spontaneous tip apex changes, and the difficulty of modeling tip-atom interactions make it challenging to select manipulation parameters that can achieve atomic precision throughout extended operations.

…”