1.1 TMACS/mW Fine-Grained Stochastic Resonant Charge-Recycling Array Processor

Karakiewicz, R.; Genov, Roman; Cauwenberghs, Gert

doi:10.1109/jsen.2011.2113393

Cited by 25 publications

(14 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The numerical results suggest that 2-layer BMNNs can work just as well as 2-layer RMNN, although they may require a larger width. The weights of the BMNNs we have trained can now be immediately implemented in a hardware chip, such as (Karakiewicz et al, 2012), significantly improving their speed and energy efficiency in comparison to software-based RMNNs. It remains to be seen whether deep BMNNs can compete with RMNNs with (usually, fine tuned) deep architectures, which achieve state-of-the-art performance.…”

Section: Discussionmentioning

confidence: 99%

“…For example, it could be very useful if weights are restricted to assume only binary values (e.g., ±1). This may allow a dense, fast and energetically efficient hardware implementation of MNNs (e.g., with the chip in (Karakiewicz et al, 2012), which can perform 10 12 operations per second with 1mW power efficiency). Limiting the weights to binary values only mildly reduces the (linear) computational capacity of a MNN (at most, by a logarithmic factor (Ji & Psaltis, 1998)).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Mean Field Bayes Backpropagation: scalable training of multilayer neural networks with binary weights

Soudry¹,

Meir²

2013

Preprint

View full text Add to dashboard Cite

Significant success has been reported recently ucsing deep neural networks for classification. Such large networks can be computationally intensive, even after training is over. Implementing these trained networks in hardware chips with a limited precision of synaptic weights may improve their speed and energy efficiency by several orders of magnitude, thus enabling their integration into small and low-power electronic devices. With this motivation, we develop a computationally efficient learning algorithm for multilayer neural networks with binary weights, assuming all the hidden neurons have a fan-out of one. This algorithm, derived within a Bayesian probabilistic online setting, is shown to work well for both synthetic and real-world problems, performing comparably to algorithms with real-valued weights, while retaining computational tractability.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Mean Field Bayes Backpropagation: scalable training of multilayer neural networks with binary weights

Soudry¹,

Meir²

2013

Preprint

View full text Add to dashboard Cite

show abstract

“…Recently, a 10 15 operations/watt analog deep machinelearning engine composed of an 8x4 array of parallel reconfigurable analog computation cells (RAC) was presented in [19], which mimics the hierarchical presentation of information in the human brain to achieve robust automated feature extraction with the accuracy comparable to the baseline software simulations. In [20], a mixed-signal VLSI array with 1.1 TMACS (10 12 multiply-and-accumulates per second) per mW is presented, which is used in applications like pattern recognition and data compression.…”

Section: Analog Image and Video Processingmentioning

confidence: 99%

Understanding the landscape of accelerators for vision

Chandramoorthy

Swaminathan

Cotter

et al. 2014

2014 IEEE Workshop on Signal Processing Systems (SiPS)

View full text Add to dashboard Cite

Visual analytics applications are becoming ubiquitous and embedded in various systems that we interact with daily. Limited power budgets and the need for high performance for cognitive visual analytics have led to a three-pronged approach of integrating advances in algorithms, architectures and technology towards designing next generation vision accelerators. Vision applications benefit from increasing processor customization, emerging devices and technologies such as Tunnel-FETs and Resistive-RAMs, and trends in non-Boolean computing such as Cellular Neural Networks (CNNs) and neuromorphic architectures. This paper provides an overview of the evolving landscape of vision accelerators.

show abstract

“…Vector–vector or matrix–vector multiplication (VVM/MVM) are pervasive operations and the basis of the most relevant and commonly used algorithms in signal processing: 1,2 fast Fourier transforms (FFTs), convolutions, digital filters, and neural networks are some of the most prominent examples. In particular, deep and convolutional neural networks (abbreviated as DNN and ConvNets, respectively), are nowadays receiving considerable attention in the area of machine learning due to its high efficacy in classification tasks.…”

Section: Introductionmentioning

confidence: 99%

Simplicial computation: A methodology to compute vector–vector multiplications with reduced complexity

Julián

Andreou

Villemur

et al. 2021

Circuit Theory & Apps

View full text Add to dashboard Cite

Summary In this paper, we propose the use of the simplicial algorithm, originally proposed to implement piecewise‐linear functions, to compute a digital vector–vector multiplication (VVM) without multiplications. The main contributions of the proposed methodology are (a) an improved error propagation with respect to parameter quantization; (b) a more efficient digital implementation with respect to area, energy, and speed in the case of large number of inputs; and (c) an intrinsic capability to produce multivariable nonlinear processing. We show that when quantization of inputs and parameters are considered, the simplicial method achieves the same accuracy with fewer representation bits for the parameters, assuming the same quantization for the inputs. Actually, in the particular case of a large number of inputs, the simplicial method needs half the number of parameter bits of a linear combination plus one.

show abstract

1.1 TMACS/mW Fine-Grained Stochastic Resonant Charge-Recycling Array Processor

Cited by 25 publications

References 23 publications

Mean Field Bayes Backpropagation: scalable training of multilayer neural networks with binary weights

Mean Field Bayes Backpropagation: scalable training of multilayer neural networks with binary weights

Understanding the landscape of accelerators for vision

Simplicial computation: A methodology to compute vector–vector multiplications with reduced complexity

Contact Info

Product

Resources

About