NIST researchers are exploring new approaches to artificial intelligence (AI) by designing AI systems from the hardware level, rather than relying on traditional algorithmic approaches. The goal is to let the physics of novel devices do the computational work.
Two types of devices are being investigated:
1. Hybrid magnetic-superconducting devices that mimic biological neurons, producing voltage spikes 8 orders of magnitude faster than biological neurons
2. Magnetic tunnel junctions that could exploit thermal energy for probabilistic computations
The team has demonstrated hybrid magnetic-superconducting devices that can serve as artificial synapses in superconducting bio-inspired computational systems, operating at extremely low energy levels (sub-attojoule). They have also investigated using magnetic devices as spin-torque oscillator arrays for non-Boolean computations.
Modeling efforts using superconducting SPICE simulations have shown that these neural networks could potentially run at speeds over 100 GHz. The team has found no fundamental limitations on fan-out levels and digital communications, suggesting these architectures could scale with advances in Josephson junction fabrication.
The research team consists of experts from NIST and various universities and research institutions.
Source: https://www.nist.gov/programs-projects/emerging-hardware-artificial-intelligence
Keywords: Device, Josephson, Superconducting, Quantum, Computation
