This article proposes a general-purpose hybrid in-/near-memory compute SRAM (CRAM) that combines an 8T transposable bit cell with vector-based, bit-serial in-memory arithmetic to accommodate a wide range of bit-widths, from single to 32 or 64 bits, as well as a complete set of operation types, including integer and floating-point addition, multiplication, and division. This approach provides the flexibility and programmability necessary for evolving software algorithms ranging from neural networks to graph and signal processing. The proposed design was implemented in a small Internet of Things (IoT) processor in the 28-nm CMOS consisting of a Cortex-M0 CPU and 8 CRAM banks of 16 kB each (128 kB total). The system achieves 475-MHz operation at 1.1 V and, with all CRAMs active, produces 30 GOPS or 1.4 GFLOPS on 32-bit operands. It achieves an energy efficiency of 0.56 TOPS/W for 8-bit multiplication and 5.27 TOPS/W for 8-bit addition at 0.6 V and 114 MHz.
This paper mainly studies the hourly water demand forecasting performances of water supply system in shanghai with LS-SVM. The teaching-learning-based optimization (TLBO) is adopted to adjust the hyper-parameters of least squares support vector machine (LS-SVM). To improve the forecast accuracy, An ameliorated TLBO algorithm called ATLBO is introduced. The experimental results show that the model of water demand forecasting with ATLBO has better regression precision than grid search, particle swarm optimization (PSO) and TLBO.
14 December 2020 in IEEE Journal of Solid-State Circuits
We propose an ultra-low-power (ULP) image signal processor (ISP) that performs on-the-fly in-processing frame compression/decompression and hierarchical event recognition to exploit the temporal and spatial sparsity in an image sequence. This approach reduces energy consumption spent processing and transmitting unimportant image data to achieve a 16x imaging system energy gain in an intruder detection scenario. The ISP was fabricated in 40-nm CMOS and consumes only 170 μW at 5 frames/s for neural network-based intruder detection and 192x compressed image recording.
9 December 2020 in IEEE Solid-State Circuits Letters
A stacked voltage domain SRAM is proposed as an effective leakage reduction technique where bit-cell arrays are split into two voltage domains (top and bottom) connecting in series between VDD and GND to generate a sub-threshold retention voltage directly from a nominal supply with no area penalty or efficiency loss compared to the conventional voltage regulator approach. The Zigzag 8T bit-cell structure is chosen with an optimal transistor sizing to balance among hold stability, leakage, and area density. SRAM peripherals remains at full supply domain resulting in super-cutoff read for improved sensing margin and word-line overdrive for better write margin. A novel array swapping mechanism with a comprehensive timing control ensure stable access to arbitrary arrays within one system clock cycle. The proposed SRAM achieves 1.03pW/b leakage at 0.58V in 40nm.
This paper presents the Neural Cache architecture, which re-purposes cache structures to transform them into massively parallel compute units capable of running inferences for Deep Neural Networks. Techniques to do in-situ arithmetic in SRAM arrays, create efficient data mapping and reducing data movement are proposed. The Neural Cache architecture is capable of fully executing convolutional, fully connected, and pooling layers in-cache. The proposed architecture also supports quantization in-cache. Our experimental results show that the proposed architecture can improve inference latency by 18.3x over state-of-art multi-core CPU (Xeon E5), 7.7x over server class GPU (Titan Xp), for Inception v3 model. Neural Cache improves inference throughput by 12.4x over CPU (2.2x over GPU), while reducing power consumption by 50% over CPU (53% over GPU).