Circuit designs for superconducting optoelectronic loop neurons (1805.01947v2)

Abstract: Optical communication achieves high fanout and short delay advantageous for information integration in neural systems. Superconducting detectors enable signaling with single photons for maximal energy efficiency. We present designs of superconducting optoelectronic neurons based on superconducting single-photon detectors, Josephson junctions, semiconductor light sources, and multi-planar dielectric waveguides. These circuits achieve complex synaptic and neuronal functions with high energy efficiency, leveraging the strengths of light for communication and superconducting electronics for computation. The neurons send few-photon signals to synaptic connections. These signals communicate neuronal firing events as well as update synaptic weights. Spike-timing-dependent plasticity is implemented with a single photon triggering each step of the process. Microscale light-emitting diodes and waveguide networks enable connectivity from a neuron to thousands of synaptic connections, and the use of light for communication enables synchronization of neurons across an area limited only by the distance light can travel within the period of a network oscillation. Experimentally, each of the requisite circuit elements has been demonstrated, yet a hardware platform combining them all has not been attempted. Compared to digital logic or quantum computing, device tolerances are relaxed. For this neural application, optical sources providing incoherent pulses with 10,000 photons produced with efficiency of 10$^{-3}$ operating at 20\,MHz at 4.2\,K are sufficient to enable a massively scalable neural computing platform with connectivity comparable to the brain and thirty thousand times higher speed.

• The paper presents innovative circuits that integrate SPDs, JJs, and photonic signals to emulate biological neural networks with minimal energy consumption.
• The design employs spike-timing-dependent plasticity with single-photon synaptic weight modulation, enabling precise and efficient neuronal operations.
• Simulations show scalability with die-level networks using ~1 mW for 8,100 neurons and wafer-scale systems operating within one watt, promising neuromorphic advancements.

Circuit Designs for Superconducting Optoelectronic Loop Neurons

In this paper, the authors present innovative designs for superconducting optoelectronic neurons, leveraging advanced components such as superconducting single-photon detectors (SPDs), Josephson junctions (JJs), semiconductor light sources, and dielectric waveguides. The primary motivation hinges on the ability to integrate optical communication with superconducting electronics, achieving high energy efficiency and scalability that could emulate biological neural systems.

Overview of Superconducting Optoelectronic Neuron Design

The design utilizes superconducting SPDs to detect single-photon signals, transducing these photonic communication events into supercurrent stored within a superconducting loop. This storage is analogously similar to the integration of synaptic signals in biological neurons. The critical task of synaptic weight modulation is addressed via dynamic adjustments using superconducting circuits. A notable feature is spike-timing-dependent plasticity (STDP), implemented with minimal energy investment by employing facilities capable of updating synaptic weights with single photons.

A fundamental aspect of these neurons is the ability to achieve complex synaptic and neuronal operations while maintaining high energy efficiency. Notably, the circuits facilitate the sending of few-photon signals to numerous synaptic connections. Resultantly, neurons can synchronize over vast areas, determined only by the reach of light within the oscillation period of the network.

Strong Numerical Results

The analysis provides substantive insights into the zero resistive state of superconducting devices, effectively managing to produce neuronal firing events with minimal energy consumption. The circuits, composed of SPDs and JJs, achieve synaptic independent signaling, which is pivotal in ensuring the linear addition of signals from various synapses.

The system simulations suggest that die-level networks with approximately 8,100 neurons dissipate merely one milliwatt of device power, whilst wafer-scale networks with one million neurons operate within an attainable energy budget of one watt. These results underscore the feasibility of utilizing superconducting optoelectronic neurons in scalable, massively interconnected neural computing platforms.

Implications and Prospects

Practically, the proposed hardware platform offers a promising direction for future neuromorphic computing, especially for applications requiring high connectivity and energy efficiency, such as image processing under faint-light conditions. Theoretically, it introduces a paradigm capable of immense computational speed and efficiency unprecedented in biological analogs, potentially achieving thirty thousand times the operational speed of the human brain.

Moreover, integrating such technology with existing systems—cryogenic sensors for astronomical or medical imaging, for instance—could revolutionize data processing and transmission methodologies. There lies notable potential in hybrid implementations, harmonizing elements of superconducting technologies with conventional CMOS systems to capitalize on the strengths of both domains.

Future Developments

The prospects for superconducting optoelectronic neural platforms are expansive, possibly influencing domains such as hybrid computing, neural interfaces, and quantum systems. These designs are a precursor to cognitive systems that not only emulate biological complexities but do so with exceptional scalability and efficiency. Going forward, further exploration into optimizing light source integration, reduction of power dissipation, and practical implementation strategies will be pivotal in realizing the full scope and capabilities suggested by the authors.

The outlined architecture and principles position superconducting optoelectronic circuits at the forefront of developing future neuromorphic systems, adding a compelling dimension to our understanding and utilization of artificial intelligence systems.