Figure - available from: Neuromorphic Computing and Engineering
This content is subject to copyright. Terms and conditions apply.
Adaptivity of JSCC to write-verified 1 M-cell 1T1R RRAM array (A) Write-verified RRAM channel characteristics. GT is target conductance representing the midpoint values of desired ranges (range bounds in green dotted line). GM is corresponding conductance distributions read 1 min after programming plotted as 1-standard deviation error bar. Rate-distortion curves measured with PSNR (B) and with SSIM (C), averaged over 100 64 × 64 pixel test natural images: Write-verified RRAM analog storage with JSCC optimized for MSE (results acquired at ∼ 1 min (dark blue) and ∼ 15 h (light blue) after programming show indistinguishable difference, and therefore the two curves overlap each other) improves over one-shot programmed PCM analog storage with JSCC optimized for MSE (red, inserted figure) because of smaller device variation; Write-verified RRAM digital storage with JPEG (black), JPEG 2000 (blue), and WEBP (orange). Ideal write-verified RRAM transmission rate (dashed line) is 3.82 bits/device, realistic write-verified RRAM transmission rate (solid line) is 2.85 bits/device. JSCC consistently requires less devices/pixel than conventional image codecs to achieve the same PSNR/SSIM index. (D) Example images compressed by different methods: JSCC+ analog write-verified RRAM, JSCC+ analog one-shot programmed PCM, and WEBP+ 3.82 bits/device digital write-verified RRAM.
Source publication
Data stored in the cloud or on mobile devices reside in physical memory systems with finite sizes. Today, huge amounts of analog data, e.g. images and videos, are first digitalized and then compression algorithms (e.g., the JPEG standard) are employed to minimize the amount of physical storage required. Emerging non-volatile-memory (NVM) technologi...
Citations
... Utilizing memristors for computing and storage in the same physical unit can fully leverage this advantage. Recent studies have reported using memristor arrays to experimentally demonstrate JPEG-based image compression 34,35 and compressed data storage 36,37 . However, no study has been conducted on fully memristor-based storage in which the data compression, decompression, and storage are fully hardware implemented based on the memristor arrays. ...
... In Table 1, we summarized and compared the reported implementations of emerging NVM-based image compression techniques and storage after compression. Previous studies focused mainly on using memristors to only implement the frequency domain transformation step in traditional compression algorithms 34,35 , or using phase change memory (PCM)/memristors only for storing compressed data 36,37,45 . Meanwhile, the restored image quality after decompression is not very high. ...
The exponential growth of various complex images is putting tremendous pressure on storage systems. Here, we propose a memristor-based storage system with an integrated near-storage in-memory computing-based convolutional autoencoder compression network to boost the energy efficiency and speed of the image compression/retrieval and improve the storage density. We adopt the 4-bit memristor arrays to experimentally demonstrate the functions of the system. We propose a step-by-step quantization aware training scheme and an equivalent transformation for transpose convolution to improve the system performance. The system exhibits a high (>33 dB) peak signal-to-noise ratio in the compression and decompression of the ImageNet and Kodak24 datasets. Benchmark comparison results show that the 4-bit memristor-based storage system could reduce the latency and energy consumption by over 20×/5.6× and 180×/91×, respectively, compared with the server-grade central processing unit-based/the graphics processing unit-based processing system, and improve the storage density by more than 3 times.
... Recently, engineers have innovated memory storage with the introduction of analog, nonvolatile ferroelectric field-effect [102,103], resistive random access memory [104][105][106], magnetic random access memory [107,108] and phase change memory technologies [109][110][111]. Analog, non-volatile memory has been instrumental in the continuing maturation of AI-based neural networks [84,112,113], image analytic platforms [114] and bio-sensor devices [115,116]. ...
Over the last 75 years, artificial intelligence has evolved from a theoretical concept and novel paradigm describing the role that computers might play in our society to a tool with which we daily engage. In this review, we describe AI in terms of its constituent elements, the synthesis of which we refer to as the AI Silecosystem. Herein, we provide an historical perspective of the evolution of the AI Silecosystem, conceptualized and summarized as a Kuhnian paradigm. This manuscript focuses on the role that the AI Silecosystem plays in oncology and its emerging importance in the care of the community oncology patient. We observe that this important role arises out of a unique alliance between the academic oncology enterprise and community oncology practices. We provide evidence of this alliance by illustrating the practical establishment of the AI Silecosystem at the City of Hope Comprehensive Cancer Center and its team utilization by community oncology providers.
We hereby present experimental demonstration of the energy-efficient ferroelectric based electrostatically doping (Fe-ED) reconfigurable field-effect transistors (RFETs). By incorporating the ferroelectric program gates (Fe-PGs), our Fe-ED RFETs obtain the non-volatile reconfigurability between N/P modes. This achievement leads to an unprecedentedly low energy consumption per programming event, measuring below the nanojoule range. Additionally, these devices exhibit remarkable endurance, with-standing over 10
<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">4</sup>
programmable cycles even under 1 ms pulse stress, and acceptable retention, maintaining their operation modes for at least 10
<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">4</sup>
s. This device strategy represents a significant leap toward the development of highly energy-efficient reconfigurable integrated circuits.
