Hardware Acceleration Implementation of Sparse Coding Algorithm With Spintronic Devices
- 16 May 2019
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Nanotechnology
- Vol. 18 (1536125X) , 518-531
Abstract
Sparse coding provides a powerful means to perform feature extraction on high-dimensional data and recently has attracted broad interest for applications. However, it still suffers from a big challenge to realize real-time energy-efficient feature extraction with conventional CPUs/GPUs owing to its highly intensive computation. Benefitting from the non-volatility, low power, high speed, great scalability and good compatibility with CMOS technology, spintronic devices, such as magnetic tunnel junction (MTJ) and domain wall motion (DWM) device, have been explored for applications, ranging from memory, logic, to neuromorphic computing. In this paper, we explore the possibility of hardware acceleration implementation of sparse coding algorithm with spintronic devices by a series of design optimizations across the algorithm, architecture, circuit and device. Firstly, one sparse coding algorithm is selected well-suited for parallelization and hardware acceleration. Then, a DWM based compound spintronic device (CSD) is engineered and modelled, which is envisioned to achieve multiple conductance states. Sequentially, a parallel architecture is presented based on a dense cross-point array of the proposed DWM based CSD, which can be envisioned to accelerate the selected sparse coding algorithm with a designed dedicated periphery read and write circuit owing to its massively parallel read and write operation. Experimental results show that the selected sparse coding algorithm can be accelerated by 1400× with the proposed parallel architecture in comparison with software implementation. Moreover, its energy dissipation is 8 orders of magnitude smaller than that with software implementation. Additionally, an artificial neural circuit (ANC) with the proposed DWM based CSD is also presented, which can achieve a multi-step transfer function. By using the proposed DWM based CSD as a single synapse and the proposed ANC as one neuron, a fully-forward-connected artificial neural network (ANN) is constructed to classify the learned feature vectors. And the recognition error rate can reach 7.25%.Keywords
Funding Information
- National Postdoctoral Program for Innovation Talents (BX20180028)
- China Postdoctoral Science Foundation (2018M641153)
- National Science Foundation of China Projects (61571023, 11201020)
- International Collaboration (B16001)
- National Key Technology Program of China (2017ZX01032101)
- Special Foundation of Beijing Municipal Science and Technology Commission (Z161100000216149)
This publication has 50 references indexed in Scilit:
- Vertical-current-induced domain-wall motion in MgO-based magnetic tunnel junctions with low current densitiesNature Physics, 2011
- LIBSVMACM Transactions on Intelligent Systems and Technology, 2011
- A perpendicular-anisotropy CoFeB–MgO magnetic tunnel junctionNature Materials, 2010
- High Domain Wall Velocities due to Spin Currents Perpendicular to the PlanePhysical Review Letters, 2009
- The emergence of spin electronics in data storageNature Materials, 2007
- $rm K$-SVD: An Algorithm for Designing Overcomplete Dictionaries for Sparse RepresentationIEEE Transactions on Signal Processing, 2006
- Micromagnetic understanding of current-driven domain wall motion in patterned nanowiresEurophysics Letters, 2005
- An iterative thresholding algorithm for linear inverse problems with a sparsity constraintCommunications on Pure and Applied Mathematics, 2004
- Least angle regressionThe Annals of Statistics, 2004
- Gradient-based learning applied to document recognitionProceedings of the IEEE, 1998