Energy-efficient stochastic computing with superparamagnetic tunnel junctions

Abstract

Superparamagnetic tunnel junctions (SMTJs) have emerged as a competitive, realistic nanotechnology to support novel forms of stochastic computation in CMOS-compatible platforms. One of their applications is to generate random bitstreams suitable for use in stochastic computing implementations. We describe a method for digitally programmable bitstream generation based on pre-charge sense amplifiers. This generator is significantly more energy efficient than SMTJ-based bitstream generators that tune probabilities with spin currents and a factor of two more efficient than related CMOS-based implementations. The true randomness of this bitstream generator allows us to use them as the fundamental units of a novel neural network architecture. To take advantage of the potential savings, we codesign the algorithm with the circuit, rather than directly transcribing a classical neural network into hardware. The flexibility of the neural network mathematics allows us to adapt the network to the explicitly energy efficient choices we make at the device level. The result is a convolutional neural network design operating at $\approx$ 150 nJ per inference with 97 % performance on MNIST -- a factor of 1.4 to 7.7 improvement in energy efficiency over comparable proposals in the recent literature.

arXiv comment20 pages (12 pages main text), 12 figures

References in corpus (18)

• DoReFa-Net: Training Low Bitwidth Convolutional Neural Networks with Low Bitwidth Gradients
• Quantized Neural Networks: Training Neural Networks with Low Precision Weights and Activations
• Deep Learning with Limited Numerical Precision
• Stochastic p-bits for Invertible Logic
• VLSI Implementation of Deep Neural Network Using Integral Stochastic Computing
• p-Bits for Probabilistic Spin Logic
• Magnetoresistance and spin-transfer torque in magnetic tunnel junctions
• Low-Energy Truly Random Number Generation with Superparamagnetic Tunnel Junctions for Unconventional Computing
• Probabilistic Deep Spiking Neural Systems Enabled by Magnetic Tunnel Junction
• Hardware emulation of stochastic p-bits for invertible logic
• Low Barrier Nanomagnets as p-bits for Spin Logic
• Adaptive learning by extremal dynamics and negative feedback
• Implementing Bayesian Networks with Embedded Stochastic MRAM
• Thermal Relaxation Rates of Magnetic Nanoparticles in the Presence of Magnetic Fields and Spin-Transfer Effects
• Low Barrier Magnet Design for Efficient Hardware Binary Stochastic Neurons
• SC-DCNN: Highly-Scalable Deep Convolutional Neural Network using Stochastic Computing
• Voltage-driven Building Block for Hardware Belief Networks
• Reservoir Computing using Stochastic p-Bits

Cited by in corpus (21)

• Computing and Memory Technologies based on Magnetic Skyrmions
• Quantum materials for energy-efficient neuromorphic computing
• Stochastic Computing Implemented by Skyrmionic Logic Devices
• Autonomous Probabilistic Coprocessing with Petaflips per Second
• Quantitative Evaluation of Hardware Binary Stochastic Neurons
• Double Free-Layer Magnetic Tunnel Junctions for Probabilistic Bits
• Periodic structure of memory function in spintronics reservoir with feedback current
• Mutual control of stochastic switching for two electrically coupled superparamagnetic tunnel junctions
• Nanomagnetic Self-Organizing Logic Gates
• Analog Signal Processing Using Stochastic Magnets
• Double-Free-Layer Stochastic Magnetic Tunnel Junctions with Synthetic Antiferromagnets
• Voltage-insensitive stochastic magnetic tunnel junctions with double free layers
• Chaotic Antiferromagnetic Nano-Oscillator driven by Spin-Torque
• Superior probabilistic computing using operationally stable probabilistic-bit constructed by manganite nanowire
• Comparison of measured and simulated spin-wave mode spectra of magnetic nanostructures
• Improved circuits for a biologically-inspired random pulse computer
• Measurement-driven neural-network training for integrated magnetic tunnel junction arrays
• Magnetic tunnel junction as a real-time entropy source: Field-Programmable Gate Array based random bit generation without post-processing
• Quantized Artificial Neural Networks Implemented with Spintronic Stochastic Computing
• 2022 Roadmap on Neuromorphic Computing and Engineering
• Building Reservoir Computing Hardware Using Low Energy-Barrier Magnetics