In-Memory Associative Chip for NLP

Baden A

Question Answering (QA) is an area of great interest in natural language processing (NLP).

This presentation introduces an In-Memory associative processing unit (APU) chip architecture that is well-suited to processing large corpora and can be applied to QA and various NLP tasks such as language translation.

The APU chip  is a fully programmable  associative, content-based, retrieval and compute in-memory architecture that can embed long documents and compute in-place (inside the associative memory) any type of memory network and answer complex questions in O(1).

Speakers
Avidan Akerib
Avidan
Akerib
Vice President Associative Computing
at
GSI Technology

Dr. Avidan Akerib

Summary

30 years of experience in parallel computing and In Place Associative Computing. Over 25 Granted Patents related to Parallel and in-memory associative computing Technology. PhD, Applied mathematics & Computer Science from Weismann Inst of science, Israel

Specialties
Computational Memory
Associative Processing
Parallel Algorithms
Machine Learning

Work Experience

• Nov 2015- Current: Vice President Associative Computing BU , GSI Technology.
Managing a new BU related to a In-Place Associative Computing products for Big Data and Cyber Security Applications, based on GSI Acquisition of MikaMonu Group Ltd.
• 2008 – Nov 2015t: MikaMonu Group Ltd: Cofounder and CTO
Mikamonu Group Develops an Innovative technology and products with very high performance and low power (from 100X-1000X better than standard solutions) for applications in cloud computing image processing and storage. The technology is based on "converting" commercial Memory such as SRAM to highly parallel computing where processing is implemented inside the memory. The products include SW tools, Evaluation boards and VLSI chips

• 2008 – 2011: ZikBit Ltd: Cofounder and CTO
ZikBit developed an innovative technology that is implemented on DRAM memory chips. The innovative memory chips will enable an ultra rich user experience in mobile devices while cutting cost and power.

• 2007 – 2008 OrYashir Ltd: Cofounder and CEO
Consultant for "Bright Scale" ( US Company) in the implantation of HDTV codec and Frame rate conversions on parallel SISD processor. The profit from this startup was used to invent new patents in the field of computational DRAM. The IP was sold to ZikBit investors.

• 2001 – 2007 General Manager Neomagic Ltd, Vice President Neomagic Inc.
As technology leader of 25 employees in Israel and 10 employees in India, Neomagic Israel developed the associative processing array, (APA) which was the core business of Neomagic in application processors for mobile phones and mobile TV. The APA is a programmable low power, low cost engine including 512 processors that enables Video Encode decode and 3D graphics. During this work I lead the innovation from concept to real products. The structure of the team was as follows: SW (Video/Audio/3D APIs), OS, Parallel Algorithms (Compression, Image stabilizer, Image processing, OpenGL ES, OFDM etc), Architecture, Innovation, QA, and administration. I worked very closely to the team, motivated them and reported the progress weekly to my managers at US.

• 1999 – 2001 Chief Technologist Neomagic Israel.

Development of parallel algorithms and simulations of Camera functionalities, Mpeg-4 codec for Digital Camera and Mobile phone.

• 1993 – 2001 Co Founder and President Associative Computing Ltd (ACL)
ACL was acquired by Neomagic at Feb 1999.
As co-founder and president of ACL from its inception in 1993. I led that company’s development of the innovative parallel processing XU-2 real time image processing engine, based on associative computing technology, which was used in machine vision, robotic inspection and printer applications in Israel, Japan and US Industries. The product included full custom chip, board and more than 200 functions for the robotic industry, such as pattern search, tracking, segmentation, skeletonize, color separation. ACL was subsidiary of Robomatix, a producer of laser robotic welding and cutting systems for the automotive industry.

1993 Lecturer at Technion EE Department
1991-1993 Lecturer of " introduction to Computers" at College for Management
1979-1993 Electrical eng, SW and HW at Defense Industries.
Over decade, work at several defense industries and military (8100, Tadiran and more) including the development of the first Mini Remotely Piloted Vehicle (MRPV) technologies (auto pilot, camera stabilization, control station and communication), this included the implementation of complex digital control systems on the first microprocessors such as Intel 8085 while closing the loop from sensors such as gyros and servos. I also implemented a complete image processing system for recognition, enhancement and analysis. Our team received the Israel Defense Prize after the first Lebanon war at 1982.

Education
• Phd : 1988-1992: Weizmann Institute of Science
Applied Mathematics and Computer Science
Developed the theory of associative computing and applications for image processing and graphics.

Recording