Phase-change memtransistive synapses for mixed-plasticity neural computations

Sarwat, Syed Ghazi; Kersting, Benedikt; Moraitis, Timoleon; Jonnalagadda, Vara Prasad; Sebastian, Abu

doi:10.1038/s41565-022-01095-3

Cited by 75 publications

(62 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We require analog control of memristor resistances to encode the complex weights and to control the bias currents of the individual oscillators in the case that oscillators need to be individually tuned to match their frequencies. Local analog memory using memristors is an active area of research [43,64] and our approach would not be viable without its realization. There are different constraints on the memristors we use in different parts of the network.…”

Section: Discussionmentioning

confidence: 99%

Associative memories using complex-valued Hopfield networks based on spin-torque oscillator arrays

Prasad

Mukim

Madhavan

et al. 2022

Neuromorph. Comput. Eng.

View full text Add to dashboard Cite

Simulations of complex-valued Hopﬁeld networks based on spin-torque oscillators can recover phase-encoded images. Sequences of memristor-augmented inverters provide tunable delay elements that implement complex weights by phase shifting the oscillatory output of the oscillators. Pseudo-inverse training suﬃces to store at least 12 images in a set of 192 oscillators, representing 16 × 12 pixel images. The energy required to recover an image depends on the desired error level. For the oscillators and circuitry considered here, 5 % root mean square deviations from the ideal image require approximately 5 µs and consume roughly 130 nJ. Simulations show that the network functions well when the resonant frequency of the oscillators can be tuned to have a fractional spread less than 10-3, depending on the strength of the feedback.

show abstract

Section: Discussionmentioning

confidence: 99%

Associative memories using complex-valued Hopfield networks based on spin-torque oscillator arrays

Prasad

Mukim

Madhavan

et al. 2022

Neuromorph. Comput. Eng.

View full text Add to dashboard Cite

show abstract

“…Of particular interest to us are the memristor-based synapse designs for neuromorphic computing [29][30][31][32] given the energy-efficient nature of their operation [33]. Employing materials such as Ge 15 Sb 85 and Sb on Al 2 O 3 and SiO 2 dielectrics in [34], the memtransistive synapse uses a 'phase-change' mechanism to switch between two structural states of varying conductivity. The non-volatile and reversible nature of its conductivity has been interpreted as the change of plasticity of the synapse.…”

Section: Discussionmentioning

confidence: 99%

On-device Synaptic Memory Consolidation using Fowler-Nordheim Quantum-tunneling

Rahman¹,

Bose²,

Chakrabartty³

2022

Preprint

View full text Add to dashboard Cite

Synaptic memory consolidation has been heralded as one of the key mechanisms for supporting continual learning in neuromorphic Artificial Intelligence (AI) systems. Here we report that a Fowler-Nordheim (FN) quantum-tunneling device can implement synaptic memory consolidation similar to what can be achieved by algorithmic consolidation models like the cascade and the elastic weight consolidation (EWC) models. The proposed FN-synapse not only stores the synaptic weight but also stores the synapse's historical usage statistic on the device itself. We also show that the operation of the FN-synapse is nearoptimal in terms of the synaptic life-time and we demonstrate that a network comprising FN-synapses outperforms a comparable EWC network for a small benchmark continual learning task. With an energy foot-print of femtojoules per synaptic update, we believe that the proposed FN-synapse provides an ultra-energy-efficient approch for implementing both synaptic memory consolidation and persistent learning.

show abstract

“…The 4 and 10 μm long devices have a recrystallization 100 and 200 s, respectively. The intrinsic transient behavior can be used to implement leaky integrate and fire neurons , and can be used for applications requiring combination of short- and long-term plasticity in neurons …”

Section: Programming Multiple Levelsmentioning

confidence: 99%

Antimony as a Programmable Element in Integrated Nanophotonics

et al. 2022

View full text Add to dashboard Cite

The use of nonlinear elements with memory as photonic computing components has seen a huge surge in interest in recent years with the rise of artificial intelligence and machine learning. A key component is the nonlinear element itself. A class of materials known as phase change materials has been extensively used to demonstrate the viability of such computing. However, such materials continue to have relatively slow switching speeds, and issues with cyclability related to phase segregation of phase change alloys. Here, using antimony (Sb) thin films with thicknesses less than 5 nm we demonstrate reversible, ultrafast switching on an integrated photonic platform with retention time of tens of seconds. We use subpicosecond pulses, the shortest used to switch such elements, to program seven distinct memory levels. This portends their use in ultrafast nanophotonic applications ranging from nanophotonic beam steerers to nanoscale integrated elements for photonic computing.

show abstract

Phase-change memtransistive synapses for mixed-plasticity neural computations

Cited by 75 publications

References 44 publications

Associative memories using complex-valued Hopfield networks based on spin-torque oscillator arrays

Associative memories using complex-valued Hopfield networks based on spin-torque oscillator arrays

On-device Synaptic Memory Consolidation using Fowler-Nordheim Quantum-tunneling

Antimony as a Programmable Element in Integrated Nanophotonics

Contact Info

Product

Resources

About