A Non-Volatile All-Spin Analog Matrix Multiplier: An Efficient Hardware
Accelerator for Machine Learning
Abstract
We propose and analyze a compact and non-volatile nanomagnetic
(all-spin) analog matrix multiplier performing the
multiply-and-accumulate (MAC) operation using two magnetic tunnel
junctions – one activated by strain to act as the multiplier, and the
other activated by spin-orbit torque pulses to act as a domain wall
synapse that performs the operation of the accumulator. Each MAC
operation can be performed in ~1 ns and the maximum
energy dissipated per operation is ~100 aJ. This
provides a very useful hardware accelerator for machine learning (e.g.
training of deep neural networks), solving combinatorial optimization
problems with Ising type machines, and other artificial intelligence
tasks which often involve the multiplication of large matrices. The
non-volatility allows the matrix multiplier to be embedded in powerful
non-von-Neumann architectures.