Researchers have created a programmable silicon photonic neuron capable of controlling the timing of optical spikes, a step toward more adaptable neuromorphic computing systems. A key component, the Dynamically Reconfigurable Unified Microresonator (DRUM), allows electrical control over optical modes without physically altering the device, addressing a limitation of conventional photonic neurons. This control extends to characteristics mirroring biological neurons, including firing threshold, integration time and refractory period, enabling nuanced control over information processing timing and responsiveness. The device also manipulates back-action, potentially offering greater control over communication between future photonic nodes within a network.
DRUM Chip Enables Electrical Control of Optical Spiking Directionality
Silicon photonics has advanced brain-like computation with the creation of a device offering electrical control over the direction of light signals, a capability previously limited by fixed hardware configurations. The core of this advancement is the Dynamically Reconfigurable Unified Microresonator, or DRUM, which manipulates optical modes without physical alterations to the chip’s structure. This programmability addresses a key limitation of earlier photonic neurons, where forward and reverse light paths were inextricably linked, hindering independent control of signal flow.
Adjusting optical coupling conditions directly impacts when the device spikes, generates a signal, and how rapidly it can respond to subsequent stimuli, demonstrating programmable control over the timing of a spike. This precise timing is crucial for complex neuromorphic computation, where information is processed through interconnected, spiking elements.
The DRUM chip can also perform temporal integration; successive optical pulses accumulate energy until a spiking threshold is reached, with the timing of this response dictated by the coupling phase. Beyond simply mimicking neuronal behavior, the architecture allows for manipulation of back-action, the influence of reflected signals on the system. Instead of treating these backward-propagating signals as interference, the team designed them to be an adjustable component of the photonic neuron’s operation.
Simulations of two interconnected photonic nodes revealed behaviors such as inhibition and synchronization, suggesting potential for complex network interactions. While experimental demonstration of a full network implementation remains future work, these results indicate how programmable back-action could shape communication between future photonic neurons, potentially supporting heterogeneous networks where each node possesses unique characteristics tailored to its specific role.
See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
