Photonic tensor cores for machine learning

被引:143
作者
Miscuglio, Mario [1 ]
Sorger, Volker J. [1 ]
机构
[1] George Washington Univ, Dept Elect & Comp Engn, Washington, DC 20052 USA
关键词
MATRIX MULTIPLICATION; PHOTODETECTOR;
D O I
10.1063/5.0001942
中图分类号
O59 [应用物理学];
学科分类号
摘要
With an ongoing trend in computing hardware toward increased heterogeneity, domain-specific coprocessors are emerging as alternatives to centralized paradigms. The tensor core unit has been shown to outperform graphic processing units by almost 3 orders of magnitude, enabled by a stronger signal and greater energy efficiency. In this context, photons bear several synergistic physical properties while phase-change materials allow for local nonvolatile mnemonic functionality in these emerging distributed non-von Neumann architectures. While several photonic neural network designs have been explored, a photonic tensor core to perform tensor operations is yet to be implemented. In this manuscript, we introduce an integrated photonics-based tensor core unit by strategically utilizing (i) photonic parallelism via wavelength division multiplexing, (ii) high 2 peta-operations-per-second throughputs enabled by tens of picosecond-short delays from optoelectronics and compact photonic integrated circuitry, and (iii) near-zero static power-consuming novel photonic multi-state memories based on phase-change materials featuring vanishing losses in the amorphous state. Combining these physical synergies of material, function, and system, we show, supported by numerical simulations, that the performance of this 4-bit photonic tensor core unit can be 1 order of magnitude higher for electrical data. The full potential of this photonic tensor processor is delivered for optical data being processed, where we find a 2-3 orders higher performance (operations per joule), as compared to an electrical tensor core unit, while featuring similar chip areas. This work shows that photonic specialized processors have the potential to augment electronic systems and may perform exceptionally well in network-edge devices in the looming 5G networks and beyond.
引用
收藏
页数:10
相关论文
共 41 条
[1]   OE-CAM: A Hybrid Opto-Electronic Content Addressable Memory [J].
Alkabani, Yousra ;
Miscuglio, Mario ;
Sorger, Volker J. ;
El-Ghazawi, Tarek .
IEEE PHOTONICS JOURNAL, 2020, 12 (02)
[2]   100 GHz silicon-organic hybrid modulator [J].
Alloatti, Luca ;
Palmer, Robert ;
Diebold, Sebastian ;
Pahl, Kai Philipp ;
Chen, Baoquan ;
Dinu, Raluca ;
Fournier, Maryse ;
Fedeli, Jean-Marc ;
Zwick, Thomas ;
Freude, Wolfgang ;
Koos, Christian ;
Leuthold, Juerg .
LIGHT-SCIENCE & APPLICATIONS, 2014, 3 :e173-e173
[3]  
[Anonymous], 2008, IARC PUBL, VIX, P1
[4]  
[Anonymous], 2017, ARXIV170404760CS
[5]   A Faster Parallel Algorithm for Matrix Multiplication on a Mesh Array [J].
Bae, Sung Eun ;
Shinn, Tong-Wook ;
Takaoka, Tadao .
2014 INTERNATIONAL CONFERENCE ON COMPUTATIONAL SCIENCE, 2014, 29 :2230-2240
[6]   High-Responsivity Low-Voltage 28-Gb/s Ge p-i-n Photodetector With Silicon Contacts [J].
Chen, Hong Tao ;
Verheyen, Peter ;
De Heyn, Peter ;
Lepage, Guy ;
De Coster, Jeroen ;
Absil, Philippe ;
Roelkens, Gunther ;
Van Campenhout, Joris .
JOURNAL OF LIGHTWAVE TECHNOLOGY, 2015, 33 (04) :820-824
[7]  
Choukroun Y., 2019, ARXIV190206822CSSTAT
[8]   MATRIX MULTIPLICATION VIA ARITHMETIC PROGRESSIONS [J].
COPPERSMITH, D ;
WINOGRAD, S .
JOURNAL OF SYMBOLIC COMPUTATION, 1990, 9 (03) :251-280
[9]   Progress in neuromorphic photonics [J].
de Lima, Thomas Ferreira ;
Shastri, Bhavin J. ;
Tait, Alexander N. ;
Nahmias, Mitchell A. ;
Prucnal, Paul R. .
NANOPHOTONICS, 2017, 6 (03) :577-599
[10]   Inverse-designed metastructures that solve equations [J].
Estakhri, Nasim Mohammadi ;
Edwards, Brian ;
Engheta, Nader .
SCIENCE, 2019, 363 (6433) :1333-+