SAPIENS: A 64-kb RRAM-Based Non-Volatile Associative Memory for One-Shot Learning and Inference at the Edge

Li, Haitong; Chen, Wei-Chen; Levy, Akash; Wang, Ching-Hua; Wang, Hongjie; Chen, Po‐Han; Wan, Weier; Khwa, Win-San; Chuang, Harry; Chih, Yu-Der; Chang, Meng-Fan; Wong, H.-S. Philip; Raina, Priyanka

doi:10.1109/ted.2021.3110464

Cited by 47 publications

(14 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…2FeFET 24,25 ), etc. Although these studies demonstrated good energy efficiency, they are limited to at most a few bits mismatches which have difficulties serving as attentional memory modules for scaled-up MANNs [24][25][26] .…”

Section: Tcam In Crossbars With Ability To Output Degree Of Mismatchesmentioning

confidence: 99%

“…Besides, the locality sensitive hashing (LSH) function that enables the degree of mismatch search in the TCAM was implemented in software, and the experimental demonstration was limited to a 2×2 TCAM array. More recently, a 2T-2R TCAM associative memory was used to demonstrate few-shot learning by calculating L 1 distance 26 . In this work a 2-bit readout scheme is employed (requiring 64 cycles per row) which incurs high energy and latency overheads, and feature extraction is again relegated to a digital processor.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Experimentally realized memristive memory augmented neural network

Mao¹,

Wen²,

Zhao³

et al. 2022

Preprint

View full text Add to dashboard Cite

Lifelong on-device learning is a key challenge for machine intelligence, and this requires learning from few, often single, samples. Memory augmented neural network has been proposed to achieve the goal, but the memory module has to be stored in an off-chip memory due to its size. Therefore its practical use has been heavily limited. Previous works on emerging memory-based implementation had difficulties in scaling up because different modules with various structures were difficult to integrate on the same chip and the small sense margin of the content addressable memory for the memory module heavily limited the degree of mismatch calculation. In this work, we experimentally validated that all those different structures in the memory augmented neural network architecture can be implemented in a fully integrated memristive crossbar platform and achieve an accuracy that closely matches standard software on digital hardware for the Omniglot dataset. The successful demonstration is supported by implementing new functions in crossbars in addition to widely reported matrix multiplications. For example, the locality-sensitive hashing operation is implemented in crossbar arrays by exploiting the intrinsic stochasticity of memristor devices. Besides, the content-addressable memory module is realized in crossbars, which also supports the degree of mismatches. Simulations based on experimentally validated models show such an implementation can be efficiently scaled up for one-shot learning on the Mini-ImageNet dataset. The successful demonstration paves the way for practical on-device lifelong learning and opens possibilities for novel attention-based algorithms not possible in conventional hardware.

show abstract

Section: Tcam In Crossbars With Ability To Output Degree Of Mismatchesmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Experimentally realized memristive memory augmented neural network

Mao¹,

Wen²,

Zhao³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…CAMs/TCAMs based on non-volatile memories have been developed recently, including those based on memristor/ReRAM (e.g. 2T-2R 26,42 , 2.5T-1R 43 ), floating gate transistor (e.g. 2Flash 44 ), ferroelectric transistors (e.g.…”

Section: Tcam In Crossbars With Ability To Output Degree Of Mismatchesmentioning

confidence: 99%

“…2FeFET 24,25 ), etc. Although these studies demonstrated good energy efficiency, they are limited to at most a few bits mismatches which have difficulties to serve as attentional memory modules for scaled-up MANNs [24][25][26] .…”

Section: Tcam In Crossbars With Ability To Output Degree Of Mismatchesmentioning

confidence: 99%

“…Besides, the locality sensitive hashing (LSH) function that enables the degree of mismatch search in the TCAM was implemented in software, and the experimental demonstration was limited to a 2×2 TCAM array. More recently, a 2T-2R TCAM associative memory was used to demonstrate few-shot learning by calculating L 1 distance 26 . In this work a 2-bit readout scheme is employed (requires 64 cycles per row) which incurs high energy and latency overheads, and feature extraction is again relegated to a digital processor.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Experimentally realized memristive memory augmented neural network

Mao¹,

Wen²,

Zhao³

et al. 2022

Preprint

View full text Add to dashboard Cite

Lifelong on-device learning is a key challenge for machine intelligence, and this requires learning from few, often single, samples. Memory augmented neural network has been proposed to achieve the goal, but the memory module has to be stored in an off-chip memory due to its size. Therefore the practical use has been heavily limited. Previous works on emerging memory-based implementation have difficulties in scaling up because different modules with various structures are difficult to integrate on the same chip and the small sense margin of the content addressable memory for the memory module heavily limited the degree of mismatch calculation. In this work, we implement the entire memory augmented neural network architecture in a fully integrated memristive crossbar platform and achieve an accuracy that closely matches standard software on digital hardware for the Omniglot dataset. The successful demonstration is supported by implementing new functions in crossbars in addition to widely reported matrix multiplications. For example, the locality-sensitive hashing operation is implemented in crossbar arrays by exploiting the intrinsic stochasticity of memristor devices. Besides, the content-addressable memory module is realized in crossbars, which also supports the degree of mismatches. Simulations based on experimentally validated models show such an implementation can be efficiently scaled up for one-shot learning on the Mini-ImageNet dataset. The successful demonstration paves the way for practical on-device lifelong learning and opens possibilities for novel attention-based algorithms not possible in conventional hardware.

show abstract

Differentiable Content Addressable Memory with Memristors

Pedretti

Graves

Vaerenbergh

et al. 2022

Adv Elect Materials

View full text Add to dashboard Cite

Memristors, Flash, and related nonvolatile analog device technologies offer in‐memory computing structures operating in the analog domain, such as accelerating linear matrix operations in array structures. These take advantage of analog tunability and large dynamic range. At the other side, content addressable memories (CAM) are fast digital lookup tables which effectively perform nonlinear Boolean logic and return a digital match/mismatch value. Recently, nonvolatile analog CAMs have been presented merging analog storage and analog search operations with digital match/mismatch output. However, CAM blocks cannot easily be inserted within a larger adaptive system due to the challenges of training and learning with binary outputs. Here, a missing link between analog crossbar arrays and CAMs, namely a differentiable content addressable memory (dCAM), is presented. Utilizing nonvolatile memories that act as a “soft” memory with analog outputs, dCAM enables learning and fine‐tuning of the memory operation and performance. Four applications are quantitatively evaluated to highlight the capabilities: improved data pattern storage, improved robustness to noise and variability, reduced energy and latency performance, and an application to solving Boolean satisfiability optimization problems. The use of dCAM is envisioned as a core building block of fully differentiable computing systems employing multiple types of analog compute operations and memories.

show abstract

SAPIENS: A 64-kb RRAM-Based Non-Volatile Associative Memory for One-Shot Learning and Inference at the Edge

Cited by 47 publications

References 12 publications

Experimentally realized memristive memory augmented neural network

Experimentally realized memristive memory augmented neural network

Experimentally realized memristive memory augmented neural network

Differentiable Content Addressable Memory with Memristors

Contact Info

Product

Resources

About