Photonic Neuromorphic Computing Using GaN-Based Synaptic Transistors for Edge AI Acceleration
DOI:
https://doi.org/10.63282/3050-9262.IJAIDSML-V5I3P130Keywords:
Neuromorphic Computing, Photonic Synaptic Transistors, GaN/AlN Optoelectronic Synapse, Edge AI Acceleration, Synaptic Plasticity, Photonic Neural Networks, Compute-In-Memory, Spike-Timing-Dependent Plasticity, Resistive Random-Access Memory, Wide-Bandgap Semiconductors, Photonic Multiplexing, Ultra-Low Power ComputingAbstract
The interface between photonic computing and neuromorphic hardware has provided an emergent route toward realizing energy-efficiency artificial intelligence at the network edge. The paper describes a synthesized exploration of photonic neuromorphic computing systems using gallium nitride (GaN)-based synaptic transistors, with particular focus on how they can be applied to edge artificial intelligence (AI) acceleration. The alloy and its impurity, i.e. aluminum nitride (AlN) and its alloy with gallium, i.e. gallium nitride (GaN), are found to be uniquely suited materials to optoelectronic synaptic devices due to their wide bandgap (i.e. 3.4 eV for GaN), high electron mobility, and compatibility with ultraviolet (UV) photonic stimulation. Synthesized evidence on the current literature has shown that GaN/AlN nanolayered periodic structures are characterized by synaptic energy consumptions as low as 0.05 pJ per operation and endurance, and that photonic synaptic transistors using novel electron trapping layers are characterized by sub-0.02 pJ energy expenditure per spike [7] and photonic synaptic transistors using novel electron trapping layers exhibit energy expenditure per spike as low as sub-0.02 pJ [10]. At the system level, photonic convolutional accelerators have been shown to have throughputs of 11 TOPS (tera-operations per second) with optical inference latencies of sub-nanosecond [18], which is by far more bandwidth and energy efficient than conventional electronic counterparts. Comparative studies including neuromorphic hardware platforms, such as Intel Loihi, resistive random-access memory (RRAM)-based compute-in-memory chips, and emerging photonic architectures, show that GaN-based photonic synapses provide a promising combination of ultra-low power operation, multi-state synaptic weight retention over tens of thousands of seconds, and analog tunability over tens of thousands of discrete weight states [7].
References
[1] F. Ashtiani, A. J. Geers, and F. Aflatouni, "An on-chip photonic deep neural network for image classification," Nature, vol. 606, no. 7914, pp. 501–506, 2022, doi: 10.1038/s41586-022-04714-0.
[2] Y. Bai et al., "Photonic multiplexing techniques for neuromorphic computing," Nanophotonics, vol. 12, no. 5, pp. 795–817, 2023, doi: 10.1515/nanoph-2022-0485.
[3] D. V. Christensen et al., "2022 roadmap on neuromorphic computing and engineering," Neuromorphic Computing and Engineering, vol. 2, no. 2, Art. no. 022501, 2022, doi: 10.1088/2634-4386/ac4a83.
[4] M. Davies et al., "Advancing neuromorphic computing with Loihi: A survey of results and outlook," Proc. IEEE, vol. 109, no. 5, pp. 911–934, 2021, doi: 10.1109/JPROC.2021.3067593.
[5] J. Feldmann et al., "Parallel convolutional processing using an integrated photonic tensor core," Nature, vol. 589, no. 7840, pp. 52–58, 2021, doi: 10.1038/s41586-020-03070-1.
[6] T. Fu et al., "Photonic machine learning with on-chip diffractive optics," Nature Communications, vol. 14, no. 1, Art. no. 70, 2023, doi: 10.1038/s41467-022-35772-7.
[7] X. Han et al., "Artificial optoelectronic synapse with nanolayered GaN/AlN periodic structure for neuromorphic computing," ACS Applied Nano Materials, vol. 6, no. 10, pp. 8461–8467, 2023, doi: 10.1021/acsanm.3c00796.
[8] D. Ivanov et al., "Neuromorphic artificial intelligence systems," Frontiers in Neuroscience, vol. 16, Art. no. 959626, 2022, doi: 10.3389/fnins.2022.959626.
[9] J. Jang et al., "A learning-rate modulable and reliable TiOx memristor array for robust, fast, and accurate neuromorphic computing," Advanced Science, vol. 9, no. 24, Art. no. 2201117, 2022, doi: 10.1002/advs.202201117.
[10] T. Kim and K.-S. Yun, "Photonic synaptic transistors with new electron trapping layer for high performance and ultra-low power consumption," Scientific Reports, vol. 13, no. 1, Art. no. 12640, 2023, doi: 10.1038/s41598-023-39646-w.
[11] Q. Liu et al., "Circular polarization-resolved ultraviolet photonic artificial synapse based on chiral perovskite," Nature Communications, vol. 14, no. 1, Art. no. 7290, 2023, doi: 10.1038/s41467-023-43034-3.
[12] J.-G. Min, H. Park, and W.-J. Cho, "Milk–Ta2O5 hybrid memristors with crossbar array structure for bio-organic neuromorphic chip applications," Nanomaterials, vol. 12, no. 17, Art. no. 2978, 2022, doi: 10.3390/nano12172978.
[13] H. Seok et al., "Synapse-mimetic hardware-implemented resistive random-access memory for artificial neural network," Sensors, vol. 23, no. 6, Art. no. 3118, 2023, doi: 10.3390/s23063118.
[14] A. Totovic et al., "Programmable photonic neural networks combining WDM with coherent linear optics," Scientific Reports, vol. 12, no. 1, Art. no. 5545, 2022, doi: 10.1038/s41598-022-09370-y.
[15] W. Wan et al., "A compute-in-memory chip based on resistive random-access memory," Nature, vol. 608, no. 7923, pp. 504–512, 2022, doi: 10.1038/s41586-022-04992-8.
[16] G. Wetzstein et al., "Inference in artificial intelligence with deep optics and photonics," Nature, vol. 588, no. 7836, pp. 39–47, 2020, doi: 10.1038/s41586-020-2973-6.
[17] L. G. Wright et al., "Deep physical neural networks trained with backpropagation," Nature, vol. 601, no. 7894, pp. 549–555, 2022, doi: 10.1038/s41586-021-04223-6.
[18] X. Xu et al., "11 TOPS photonic convolutional accelerator for optical neural networks," Nature, vol. 589, no. 7840, pp. 44–51, 2021, doi: 10.1038/s41586-020-03063-0.
[19] P. Yao et al., "Fully hardware-implemented memristor convolutional neural network," Nature, vol. 577, no. 7792, pp. 641–646, 2020, doi: 10.1038/s41586-020-1942-4.
[20] H. Zhu et al., "Optically readable organic electrochemical synaptic transistors for neuromorphic photonic image processing," Nano Letters, vol. 23, no. 11, pp. 5264–5271, 2023, doi: 10.1021/acs.nanolett.3c01291.
