A Scalable Multi- TeraOPS Deep Learning Processor Core for AI Trainina and Inference

Fleischer, Bruce; Shukla, Sunil; Ziegler, Matthew M.; Silberman, J. A.; Oh, Jinwook; Srinivasan, Vijayalakshmi; Choi, Jungwook; Mueller, Silvia Melitta; Agrawal, Ankur; Babinsky, Tina; Cao, Nianwen; Chen, Chia-Yu; Chuang, Pierce; Fox, Thomas W.; Gristede, George; Guillorn, Michael A.; Haynie, Howard; Klaiber, Michael J.; Lee, Dongsoo; Lo, Shih-Hsien; Maier, Gary; Scheuermann, M.; Venkataramani, Swagath; Vezyrtzis, Christos; Wang, Naigang; Yee, Fanchieh; Zhou, Ching; Lu, Pong‐Fei; Curran, Brian; Chang, Lel; Gopalakrishnan, Kailash

doi:10.1109/vlsic.2018.8502276

Cited by 97 publications

(42 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…At the same time, new opportunities for hardware security will also be provided by approximate computing (e.g., inexact hardware has been used for hiding security information and authentication for resourceconstrained Internet-of-Things (IoT) devices). In the coming decade, it is anticipated that approximate computing techniques will likely be employed in energy-efficient systems for applications such as AI and signal processing; for example, IBM fabricated an AI accelerator chip to achieve multi-tera operations per second (TOPS) performance by applying multiple approximate techniques (including hardware, architecture, algorithms, and programs) [21]. Furthermore, new approximate computing applications will be exploited.…”

Section: Reliability and Securitymentioning

confidence: 99%

A Retrospective and Prospective View of Approximate Computing [Point of View}

2020

View full text Add to dashboard Cite

C omputing systems are conventionally designed to operate as accurately as possible. However, this trend faces severe technology challenges, such as power consumption, circuit reliability, and high performance. For nearly half a century, performance and power consumption of computing systems have been consistently improved by relying mostly on technology scaling. As per Dennard's scaling, the size of a transistor has been considerably shrunk and the supply voltage has been reduced over the years, such that circuits operate at higher frequencies but nearly at the same power dissipation level. However, as Dennard's scaling tends toward an end, it is difficult to further improve performance under the same

show abstract

Section: Reliability and Securitymentioning

confidence: 99%

A Retrospective and Prospective View of Approximate Computing [Point of View}

2020

View full text Add to dashboard Cite

show abstract

“…The first foundational progress in compute efficiency for AI model training and inference has been made by exploiting the statistical and approximate nature of deep learning algorithms, leading to reduced-precision digital approaches [7] [8]. Floating-point multiplication dominates deep learning training, and multiplier complexity is quadratic with operand size; e.g., 16-bit precision engines are more than 4x smaller than 32-bit precision engines.…”

Section: Neurons: Biology + Informationmentioning

confidence: 99%

1.4 The Future of Computing: Bits + Neurons + Qubits

Gil

Green

2020

2020 IEEE International Solid- State Circuits Conference - (ISSCC)

View full text Add to dashboard Cite

The laptops, cell phones, and internet applications commonplace in our daily lives are all rooted in the idea of zeros and ones -in bits.This foundational element originated from the combination of mathematics and Claude Shannon's Theory of Information. Coupled with the 50-year legacy of Moore's Law, the bit has propelled the digitization of our world.In recent years, artificial intelligence systems, merging neuron-inspired biology with information, have achieved superhuman accuracy in a range of narrow classification tasks by learning from labelled data. Advancing from Narrow AI to Broad AI will encompass the unification of learning and reasoning through neuro-symbolic systems, resulting in a form of AI which will perform multiple tasks, operate across multiple domains, and learn from small quantities of multi-modal input data.Finally, the union of physics and information led to the emergence of Quantum Information Theory and the development of the quantum bit -the qubit -forming the basis of quantum computers. We have built the first programmable quantum computers, and although the technology is still in its early days, these systems offer the potential to solve problems which even the most powerful classical computers cannot.The future of computing will look fundamentally different than it has in the past. It will not be based on more and cheaper bits alone, but rather, it will be built upon bits + neurons + qubits. This future will enable the next generation of intelligent mission-critical systems and accelerate the rate of science-driven discovery.

show abstract

“…x i is the input of the node i. Figure 5 (A) [14] shows DL algorithms are comprised of a spectrum of operations. Although matrix multiplication (gemm) is dominant, optimizing performance efficiency while maintaining accuracy requires the core architecture to efficiently support all of the auxiliary functions.…”

Section: Deep Learningmentioning

confidence: 99%

AI: from deep learning to in-memory computing

Lung¹

2019

Metrology, Inspection, and Process Control for Microlithography XXXIII

View full text Add to dashboard Cite

In the past few years, Artificial Intelligence (AI) has been a subject of intense media hype. Machine learning, deep learning (DL), and AI come up in countless articles, often outside of technology-minded publications. As the AI hype keeps growing, it is important to be able to recognize the signal in the noise, to tell apart world-changing developments from what are merely over-hyped press releases. This paper tries to explain how deep learning is working and how GPU (Graphic Processing Unite) can make it a reality. Finally, in-memory computing (IMC) for DL is introduced to point out future high performance and low power DL hardware development direction.

show abstract

A Scalable Multi- TeraOPS Deep Learning Processor Core for AI Trainina and Inference

Cited by 97 publications

References 0 publications

A Retrospective and Prospective View of Approximate Computing [Point of View}

A Retrospective and Prospective View of Approximate Computing [Point of View}

1.4 The Future of Computing: Bits + Neurons + Qubits

AI: from deep learning to in-memory computing

Contact Info

Product

Resources

About