Optical-electronic implementation of artificial neural network for ultrafast and accurate inference processing

With the rapid development of integrated nanophotonics technology, the circuit architecture for optical neural networks that construct neural networks based on integrated nanophotonics has been widely investigated. Since the optical neural network can complete the inference processing just by propagating the optical signal in the network, the inference processing by the optical neural network is expected more than one order of magnitude faster than the electronics-only implementation of an artificial neural network (ANN). In this paper, we first propose an optical vector-matrix multiplication (VMM) circuit using wavelength division multiplexing, which enables inference processing with ultra-wideband. Although the optical VMM circuit is extremely fast, the initial version is designed for fully connected network structures, which consume large amounts of power in laser light sources. As a solution to this power explosion, this paper especially proposes sparsely connected network structures for the optical VMM operation, which reduces the power consumed in the laser sources by three orders of magnitude without any speed and bandwidth degradation compared with the fully connected counterpart. Although the main part of ultra-fast ANN is VMM, batch normalization and activation function are integral parts for accurate inference processing in ANN. Batch normalization applied at inference processing is a technique for improving the inference accuracy of ANN. Without batch normalization and activation function, the inference accuracy of ANN may significantly degrade. In this paper, we next propose electronic circuit implementation of batch normalization and activation function, which significantly improves the accuracy of inference processing without sacrificing the speed performance of inference processing. Those two functions can be implemented based on an energy-efficient O-E-O repeater circuit. We specifically propose the electronic implementation of Exponential Linear Unit (ELU in short) as an activation function. It is known that ELU largely contributes to improving the inference accuracy of ANN as well as learning speed. Finally, in this paper, we demonstrate the ultra-fast and accurate operation of the optical-electronic ANN circuit using TensorFlow and optoelectronic circuit simulator.

[1]  Yann LeCun,et al.  The mnist database of handwritten digits , 2005 .

[2]  Qixiang Cheng,et al.  High-resolution Silicon Microring based Architecture for Optical Matrix Multiplication , 2019, 2019 Conference on Lasers and Electro-Optics (CLEO).

[3]  Ray T. Chen,et al.  Toward Hardware-Efficient Optical Neural Networks: Beyond FFT Architecture via Joint Learnability , 2021, IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems.

[4]  Yoshua Bengio,et al.  BinaryConnect: Training Deep Neural Networks with binary weights during propagations , 2015, NIPS.

[5]  Li Tian,et al.  An Area-Efficient 10-Bit Buffer-Reused DAC for AMOLED Column Driver ICs , 2020 .

[6]  Sung Kyu Lim,et al.  Impact of size effects in local interconnects for future technology nodes: A study based on full-chip layouts , 2014, IEEE International Interconnect Technology Conference.

[7]  Masaya Notomi,et al.  Femtofarad optoelectronic integration demonstrating energy-saving signal conversion and nonlinear functions , 2019, Nature Photonics.

[8]  Dirk Englund,et al.  Deep learning with coherent nanophotonic circuits , 2017, 2017 Fifth Berkeley Symposium on Energy Efficient Electronic Systems & Steep Transistors Workshop (E3S).

[9]  Sailing He,et al.  Low-loss and broadband 2 × 2 silicon thermo-optic Mach-Zehnder switch with bent directional couplers. , 2016, Optics letters.

[10]  Yu Cao Predictive Technology Model for Robust Nanoelectronic Design , 2011, Integrated Circuits and Systems.

[11]  Ryan Hamerly,et al.  Large-Scale Optical Neural Networks based on Photoelectric Multiplication , 2018, Physical Review X.

[12]  Sepp Hochreiter,et al.  Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs) , 2015, ICLR.

[13]  Weichen Liu,et al.  HolyLight: A Nanophotonic Accelerator for Deep Learning in Data Centers , 2019, 2019 Design, Automation & Test in Europe Conference & Exhibition (DATE).

[14]  Yu Cao,et al.  Design benchmarking to 7nm with FinFET predictive technology models , 2012, ISLPED '12.

[15]  M. Notomi,et al.  Ultrashort low-loss Ψ gates for linear optical logic on Si photonics platform , 2020 .

[16]  Roberto Proietti,et al.  Wavelength-Space Domain High-Throughput Artificial Neural Networks by Parallel Photoelectric Matrix Multiplier , 2020, 2020 Optical Fiber Communications Conference and Exhibition (OFC).

[17]  Xiaowen Wu,et al.  SUOR: Sectioned Undirectional Optical Ring for Chip Multiprocessor , 2014, JETC.

[18]  Masaya Notomi,et al.  An Optical Neural Network Architecture based on Highly Parallelized WDM-Multiplier-Accumulator , 2019, 2019 IEEE/ACM Workshop on Photonics-Optics Technology Oriented Networking, Information and Computing Systems (PHOTONICS).

[19]  Xuan Li,et al.  Parallel convolutional processing using an integrated photonic tensor core , 2021, Nature.

[20]  Ellen Zhou,et al.  Neuromorphic photonic networks using silicon photonic weight banks , 2017, Scientific Reports.