ELight: Enabling Efficient Photonic In-Memory Neurocomputing with Life Enhancement

Zhu, Hanqing; Gu, Jiaqi; Feng, Chenghao; Liu, Mingjie; Jiang, Zixuan; Chen, Ray T.; Pan, David Z.

doi:10.1109/asp-dac52403.2022.9712497

Cited by 3 publications

(5 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Table 3 highlights the comparison between OPCMs and EPCMs. Compared to EPCMs, OPCMs offer lower latency and power consumption (e.g., 60 to 130 pJ with 25 ns for read and write [12]), higher MLC capacity (up to 5 bits per cell [12]), and higher bandwidth (e.g., 1 GHz [53]) with a lifetime of 1e6-1e8 writes per cell [54]. However, such benefits come at the cost of using a complex electro-optic unit, re-design of read and write policies, accumulated optical loss and crosstalk upon scaling, need for silicon photonic links, and sensitivity to thermal variations.…”

Section: A Epcms and Opcmsmentioning

confidence: 99%

“…Aging of the cells over the long run leads to the degradation of obtainable transmission ranges, and consequently, the accuracy can drop severely. The work in [54] showed that with only 30% aged cells in a photonic tensor core, the accuracy of the multiplication decreases by 35%. To alleviate this critical issue, an aging-aware optimization methodology for PCMbased in-memory photonic tensor cores, called ''Elight,'' was presented in [54].…”

Section: Phase Change Materials For Photonic In-memory Computingmentioning

confidence: 99%

“…The work in [54] showed that with only 30% aged cells in a photonic tensor core, the accuracy of the multiplication decreases by 35%. To alleviate this critical issue, an aging-aware optimization methodology for PCMbased in-memory photonic tensor cores, called ''Elight,'' was presented in [54]. The work in [54] also described a write-aware training process for AI applications that reduces the number of write operations (by roughly 20 times), which is the primary cause of aging.…”

Section: Phase Change Materials For Photonic In-memory Computingmentioning

confidence: 99%

“…To alleviate this critical issue, an aging-aware optimization methodology for PCMbased in-memory photonic tensor cores, called ''Elight,'' was presented in [54]. The work in [54] also described a write-aware training process for AI applications that reduces the number of write operations (by roughly 20 times), which is the primary cause of aging. In addition, a post-training optimization procedure is provided to further reduce the number of writing operations.…”

Section: Phase Change Materials For Photonic In-memory Computingmentioning

confidence: 99%

“…Optimization approaches were proposed to reduce the write procedures when a GST cell is being used as an in-memory computing unit. In particular, the work in [54] showed such optimization by performing a write-aware training to reduce the number of unnecessary writes when updating the weights during the training. The same work also showed a post-training optimization approach based on block-matching-based training and column-based reordering to reduce the number of redundant writes-with a negligible effect on the model's output accuracy-to increase the endurance of PCM-based photonic tensor cores [54].…”

Section: B Endurancementioning

confidence: 99%

See 4 more Smart Citations

A Survey on Optical Phase-Change Memory: The Promise and Challenges

2023

View full text Add to dashboard Cite

Silicon photonics (SiPh) technology has facilitated the deployment of integrated photonics across different application domains, from ultra-fast communication in Datacom applications to energy-efficient optical computation in emerging hardware accelerators for machine learning. More recently, the integration of SiPh and phase change materials has created a unique opportunity to realize adaptable, reconfigurable, and programmable photonic platforms. In particular, the nonvolatile programmability in phase change materials has made them a promising candidate for implementing photonic memory cells and architectures. Accordingly, photonic memory systems and even in-memory photonic computing paradigms are on the rise, especially given their potential for improving data access in electronic and photonic processors. However, there are still many challenges in the design and fabrication of phase-change photonic integrated circuits, which need to be addressed. This article presents a comprehensive survey on the recent advances and challenges for the integration of phase change materials with contemporary photonic devices while focusing on the photonic memory application. In particular, we explore phase-change photonic memory from the material level to the architecture level by presenting an overview of different material-level characteristics of phase change materials with their optical, electrical, and thermal properties as well as their integration into SiPh devices and photonic memory architectures and their application for in-memory photonic computing. We also present a comparison with electronic memory and discuss open research challenges that must be addressed to further advance phase-change photonic memory towards successful integration into emerging computing systems.INDEX TERMS Phase change materials, phase-change memory, in-memory photonic computing, photonic integrated circuits, silicon photonics.

show abstract