Matches in SemOpenAlex for { <https://semopenalex.org/work/W4385644144> ?p ?o ?g. }
Showing items 1 to 53 of
53
with 100 items per page.
- W4385644144 abstract "Progress in artificial intelligence and machine learning over the past decade has been driven by the ability to train larger deep neural networks (DNNs), leading to a compute demand that far exceeds the growth in hardware performance afforded by Moore's law. Training DNNs is an extremely memory-intensive process, requiring not just the model weights but also activations and gradients for an entire minibatch to be stored. The need to provide high-density and low-leakage on-chip memory motivates the exploration of emerging non-volatile memory for training accelerators. Spin-Transfer-Torque MRAM (STT-MRAM) offers several desirable properties for training accelerators, including 3-4x higher density than SRAM, significantly reduced leakage power, high endurance and reasonable access time. On the one hand, MRAM write operations require high write energy and latency due to the need to ensure reliable switching. In this study, we perform a comprehensive device-to-system evaluation and co-optimization of STT-MRAM for efficient ML training accelerator design. We devised a cross-layer simulation framework to evaluate the effectiveness of STT-MRAM as a scratchpad replacing SRAM in a systolic-array-based DNN accelerator. To address the inefficiency of writes in STT-MRAM, we propose to reduce write voltage and duration. To evaluate the ensuing accuracy-efficiency trade-off, we conduct a thorough analysis of the error tolerance of input activations, weights, and errors during the training. We propose heterogeneous memory configurations that enable training convergence with good accuracy. We show that MRAM provide up to 15-22x improvement in system level energy across a suite of DNN benchmarks under iso-capacity and iso-area scenarios. Further optimizing STT-MRAM write operations can provide over 2x improvement in write energy for minimal degradation in application-level training accuracy." @default.
- W4385644144 created "2023-08-08" @default.
- W4385644144 creator A5061899878 @default.
- W4385644144 creator A5065766721 @default.
- W4385644144 creator A5083686495 @default.
- W4385644144 date "2023-08-03" @default.
- W4385644144 modified "2023-10-16" @default.
- W4385644144 title "Evaluation of STT-MRAM as a Scratchpad for Training in ML Accelerators" @default.
- W4385644144 doi "https://doi.org/10.48550/arxiv.2308.02024" @default.
- W4385644144 hasPublicationYear "2023" @default.
- W4385644144 type Work @default.
- W4385644144 citedByCount "0" @default.
- W4385644144 crossrefType "posted-content" @default.
- W4385644144 hasAuthorship W4385644144A5061899878 @default.
- W4385644144 hasAuthorship W4385644144A5065766721 @default.
- W4385644144 hasAuthorship W4385644144A5083686495 @default.
- W4385644144 hasBestOaLocation W43856441441 @default.
- W4385644144 hasConcept C113775141 @default.
- W4385644144 hasConcept C119599485 @default.
- W4385644144 hasConcept C127413603 @default.
- W4385644144 hasConcept C149635348 @default.
- W4385644144 hasConcept C2742236 @default.
- W4385644144 hasConcept C2994168587 @default.
- W4385644144 hasConcept C41008148 @default.
- W4385644144 hasConcept C46891859 @default.
- W4385644144 hasConcept C68043766 @default.
- W4385644144 hasConcept C9390403 @default.
- W4385644144 hasConceptScore W4385644144C113775141 @default.
- W4385644144 hasConceptScore W4385644144C119599485 @default.
- W4385644144 hasConceptScore W4385644144C127413603 @default.
- W4385644144 hasConceptScore W4385644144C149635348 @default.
- W4385644144 hasConceptScore W4385644144C2742236 @default.
- W4385644144 hasConceptScore W4385644144C2994168587 @default.
- W4385644144 hasConceptScore W4385644144C41008148 @default.
- W4385644144 hasConceptScore W4385644144C46891859 @default.
- W4385644144 hasConceptScore W4385644144C68043766 @default.
- W4385644144 hasConceptScore W4385644144C9390403 @default.
- W4385644144 hasLocation W43856441441 @default.
- W4385644144 hasOpenAccess W4385644144 @default.
- W4385644144 hasPrimaryLocation W43856441441 @default.
- W4385644144 hasRelatedWork W2006167304 @default.
- W4385644144 hasRelatedWork W2492225217 @default.
- W4385644144 hasRelatedWork W2942190539 @default.
- W4385644144 hasRelatedWork W3157848896 @default.
- W4385644144 hasRelatedWork W3186183842 @default.
- W4385644144 hasRelatedWork W3217339435 @default.
- W4385644144 hasRelatedWork W4235980920 @default.
- W4385644144 hasRelatedWork W4327808078 @default.
- W4385644144 hasRelatedWork W4376608629 @default.
- W4385644144 hasRelatedWork W611690454 @default.
- W4385644144 isParatext "false" @default.
- W4385644144 isRetracted "false" @default.
- W4385644144 workType "article" @default.