Deep Physical Informed Neural Networks for Metamaterial Design

Fang, Z.; Zhan, Justin

doi:10.1109/access.2019.2963375

Cited by 114 publications

(46 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Hence, the corresponding 3D extension is N = . As our most recent work [31], we mentioned that the activation function σ (s) = sin(π s) is more powerful than σ (s) = sin(s) that suggested in [20]. That is, σ (s) = sin(π s) can deal with much extremely case, such as high frequency PDEs' problems.…”

Section: A Exemplary Experiments For Laplace-beltrami Equation On 3d mentioning

confidence: 99%

See 1 more Smart Citation

A Physics-Informed Neural Network Framework for PDEs on 3D Surfaces: Time Independent Problems

Fang

Zhan

2020

IEEE Access

Self Cite

View full text Add to dashboard Cite

Partial differential equations (PDEs) on surfaces are ubiquitous in all the nature science. Many traditional mathematical methods has been developed to solve surfaces PDEs. However, almost all of these methods have obvious drawbacks and complicate in general problems. As the fast growth of machine learning area, we show an algorithm by using the physics-informed neural networks (PINNs) to solve surface PDEs. To deal with the surfaces, our algorithm only need a set of points and their corresponding normal, while the traditional methods need a partition or a grid on the surface. This is a big advantage for real computation. A variety of numerical experiments have been shown to verify our algorithm.

show abstract

Section: A Exemplary Experiments For Laplace-beltrami Equation On 3d mentioning

confidence: 99%

“…For the discrete-time method, we set up a PINN of 4 hidden layers, with 200 neurons each layer and q + 1 neurons for the output layer, where q stages Runge-Kutta method has been used. We choose σ(s) = sin(πs) as the activation function [14,15].…”

Section: Numerical Experimentsmentioning

confidence: 99%

A Physics-Informed Neural Network Framework for PDEs on 3D Surfaces: Time Independent Problems

Fang

Zhan

2020

IEEE Access

Self Cite

View full text Add to dashboard Cite

show abstract

“…[ 150 ] These specialized models, including the recently introduced physics‐informed neural networks (PINNs), [ 163 ] are an exciting and emerging area of research within DL (see Section 6), but are still relatively early in development with few applications to problems in electromagnetism and AEM systems. [ 164,165 ]…”

Section: Forward Modeling Of Aemsmentioning

confidence: 99%

Deep Learning the Electromagnetic Properties of Metamaterials—A Comprehensive Review

Khatib

Ren

Malof

et al. 2021

Adv Funct Materials

117

View full text Add to dashboard Cite

Deep neural networks (DNNs) are empirically derived systems that have transformed traditional research methods, and are driving scientific discovery. Artificial electromagnetic materials (AEMs)—including electromagnetic metamaterials, photonic crystals, and plasmonics—are research fields where DNN results valorize the data driven approach; especially in cases where conventional methods have failed. In view of the great potential of deep learning for the future of artificial electromagnetic materials research, the status of the field with a focus on recent advances, key limitations, and future directions is reviewed. Strategies, guidance, evaluation, and limits of using deep networks for both forward and inverse AEM problems are presented.

show abstract

“…Deep neural networks (DNNs) have achieved breakthroughs in a wide variety of artificial intelligence (AI) and machine learning (ML) applications, including image classification [1], speech recognition [2], and facial recognition [3], [4]. Their application has penetrated other disciplines, such as biology, materials science, and physics [5]- [7]. The DNN promises significant benefits for "Internet of Things" (IoT) devices at the edge of the network or edge computing.…”

Section: Introductionmentioning

confidence: 99%

10T SRAM Computing-in-Memory Macros for Binary and Multibit MAC Operation of DNN Edge Processors

2021

View full text Add to dashboard Cite

Computing-in-memory (CIM) is a promising approach to reduce latency and improve the energy efficiency of the multiply-and-accumulate (MAC) operation under a memory wall constraint for artificial intelligence (AI) edge processors. This paper proposes an approach focusing on scalable CIM designs using a new ten-transistor (10T) static random access memory (SRAM) bit-cell. Using the proposed 10T SRAM bit-cell, we present two SRAM-based CIM (SRAM-CIM) macros supporting multibit and binary MAC operations. The first design achieves fully parallel computing and high throughput using 32 parallel binary MAC operations. Advanced circuit techniques such as an input-dependent dynamic reference generator and an input-boosted sense amplifier are presented. Fabricated in 28 nm CMOS process, this design achieves 409.6 GOPS throughput, 1001.7 TOPS/W energy efficiency, and a 169.9 TOPS/mm 2 throughput area efficiency. The proposed approach effectively solves previous problems such as writing disturb, throughput, and the power consumption of an analog to digital converter (ADC). The second design supports multibit MAC operation (4-b weight, 4-b input, and 8-b output) to increase the inference accuracy. We propose an architecture that divides 4-b weight and 4-b input multiplication to four 2-b multiplication in parallel, which increases the signal margin by 16× compared to conventional 4-b multiplication. Besides, the capacitive digital-to-analog converter (CDAC) area issue is effectively addressed using the intrinsic bit-line capacitance existing in the SRAM-CIM architecture. The proposed approach of realizing four 2-b parallel multiplication using the CDAC is successfully demonstrated with a modified LeNet-5 neural network. These results demonstrate that the proposed 10T bit-cell is promising for realizing robust and scalable SRAM-CIM designs, which is essential for realizing fully parallel edge computing.INDEX TERMS computing-in-memory, static random access memory, deep neural network, machine learning, edge processor. * value when technology scaling factor is used. ** result when CONV1 and FL7 layers are implemented in the SRAM-CIM.

show abstract

Deep Physical Informed Neural Networks for Metamaterial Design

Cited by 114 publications

References 32 publications

A Physics-Informed Neural Network Framework for PDEs on 3D Surfaces: Time Independent Problems

A Physics-Informed Neural Network Framework for PDEs on 3D Surfaces: Time Independent Problems

Deep Learning the Electromagnetic Properties of Metamaterials—A Comprehensive Review

10T SRAM Computing-in-Memory Macros for Binary and Multibit MAC Operation of DNN Edge Processors

Contact Info

Product

Resources

About