In-memory Accelerators with Memristors Yuval Cassuto Koby Crammer Avinoam Kolodny Technion – EE ICRI-CI Retreat May 8, 2013 PU MEM NVM.

Slides:



Advertisements
Similar presentations
ABS Control Project Ondrej Ille Pre-bachelor Project.
Advertisements

Lesson 6. The Computer Operation Computer Operating Systems GUI vs. Command line The Microsoft Windows Family File Systems – How Computers Manage Data.
Chapter 1. Basic Structure of Computers
Thank you for your introduction.
Programmable Logic PAL, PLA.
The Performance of Polar Codes for Multi-level Flash Memories
5/4/2006BAE Analog to Digital (A/D) Conversion An overview of A/D techniques.
Numerical Analysis 1 EE, NCKU Tien-Hao Chang (Darby Chang)
Module R R RRR R RRRRR RR R R R R Technion – Israel Institute of Technology The Era of Many-Module SoC: Revisiting the NoC Mapping Problem Isask’har (Zigi)
Computer components IT skills: none IT concepts: computer components (input devices, output devices, memory, storage and CPU), information flow between.
HMM-BASED PATTERN DETECTION. Outline  Markov Process  Hidden Markov Models Elements Basic Problems Evaluation Optimization Training Implementation 2-D.
CS292 Computational Vision and Language Pattern Recognition and Classification.
Memory and Programmable Logic
Paper Title Your Name CMSC 838 Presentation. CMSC 838T – Presentation Motivation u Problem paper is trying to solve  Characteristics of problem  … u.
A linear time algorithm for the weighted lexicographic rectilinear 1-center problem in the plane Nir Halman, Technion, Israel This work is part of my Ph.D.
Semantic text features from small world graphs Jure Leskovec, IJS + CMU John Shawe-Taylor, Southampton.
Charles Kime & Thomas Kaminski © 2008 Pearson Education, Inc. (Hyperlinks are active in View Show mode) Chapter 6 – Selected Design Topics Part 4 – Programmable.
Technology Component: Computer Hardware – Part 2 Basic Components Dr. V.T. Raja Oregon State University.
Computer ArchitectureFall 2008 © November 12, 2007 Nael Abu-Ghazaleh Lecture 24 Disk IO.
1 Two ways to think about logic signals Fixed logic convention –High voltage always means 1, TRUE, Asserted –Low voltage always means 0, FALSE, Negated.
Overview The von Neumann Machine - the programmable digital computer Introducing the LC-3 Computer - A “toy” computer for us to learn from Computer machine.
1 EE384Y: Packet Switch Architectures Part II Load-balanced Switches Nick McKeown Professor of Electrical Engineering and Computer Science, Stanford University.
The central processing unit and main memory chapter 4, Exploring the Digital Domain The Development and Basic Organization of Computers.
Data Clustering (a very short introduction) Intuition: grouping of data into clusters so that elements from the same cluster are more similar to each other.
Neural Networks. Background - Neural Networks can be : Biological - Biological models Artificial - Artificial models - Desire to produce artificial systems.
0 What is a computer?  Simply put, a computer is a sophisticated electronic calculating machine that:  Accepts input information,  Processes the information.
Charles Kime & Thomas Kaminski © 2004 Pearson Education, Inc. Terms of Use (Hyperlinks are active in View Show mode) Terms of Use Lecture 13 – Programmable.
ERD and Memory Architectures Paul Franzon Department of Electrical and Computer Engineering
Chapter 3 Computer Hard ware
Chapter 10 Storage and File Structure Yonsei University 2 nd Semester, 2013 Sanghyun Park.
Lecture 9 of Advanced Databases Storage and File Structure (Part II) Instructor: Mr.Ahmed Al Astal.
Student : Andrey Kuyel Supervised by Mony Orbach Spring 2011 Final Presentation High speed digital systems laboratory High-Throughput FFT Technion - Israel.
Charles Kime & Thomas Kaminski © 2008 Pearson Education, Inc. (Hyperlinks are active in View Show mode) Chapter 6 – Selected Design Topics Part 4 – Programmable.
Coding and Algorithms for Memories Lecture 9 1.
COMPUTER SCIENCE QUESTIONS… BY JACK. WHAT IS THE CPU? The cpu is the central processing unit.
 DEFINE COMPUTER ? EXPLAIN CLASSIFICATION OF COMPUTER.  WHAT ARE INPUT AND OUTPUT DEVICES OF COMPUTER ? EXPALIN OUTPUT DEVICES.  WHAT ARE MEMORY AND.
Handover and Tracking in a Camera Network Presented by Dima Gershovich.
Logic Gates It’s Only Logical. Logic Gates Are the switches that computers and similar devices use. They hold their state until something changes. Are.
EE5393, Circuits, Computation, and Biology Computing with Probabilities 1,1,0,0,0,0,1,0 1,1,0,1,0,1,1,1 1,1,0,0,1,0,1,0 a = 6/8 c = 3/8 b = 4/8.
Visual Information Systems Recognition and Classification.
Error Correction and Partial Information Rewriting for Flash Memories Yue Li joint work with Anxiao (Andrew) Jiang and Jehoshua Bruck.
Chapter 3 Digital Logic Structures. 3-2 Combinational vs. Sequential Combinational Circuit always gives the same output for a given set of inputs  ex:
Quantum algorithms vs. polynomials and the maximum quantum-classical gap in the query model.
Linear Programming Maximize Subject to Worst case polynomial time algorithms for linear programming 1.The ellipsoid algorithm (Khachian, 1979) 2.Interior.
Copyright © 2004, Dillon Engineering Inc. All Rights Reserved. An Efficient Architecture for Ultra Long FFTs in FPGAs and ASICs  Architecture optimized.
Emerging Non-volatile Memories: Opportunities and Challenges
Social – End of topic evaluation. Topic 2 – Cognitive Psychology Lesson one – Introduction to the approach and topic.
Higher Computing Science 2016 Prelim Revision. Topics to revise Computational Constructs parameter passing (value and reference, formal and actual) sub-programs/routines,
Yue Li joint work with Anxiao (Andrew) Jiang and Jehoshua Bruck.
Digital Design Lecture 13
Learning Kernel Classifiers 1. Introduction Summarized by In-Hee Lee.
Coding and Algorithms for Memories Lecture 7 1.
Compression for Fixed-Width Memories Ori Rottenstriech, Amit Berman, Yuval Cassuto and Isaac Keslassy Technion, Israel.
Latches, Flip Flops, and Memory ECE/CS 252, Fall 2010 Prof. Mikko Lipasti Department of Electrical and Computer Engineering University of Wisconsin – Madison.
Energy-efficient Scheduling policy for collaborative execution in mobile cloud computing INFOCOM '13.
Flash Memory by Matt & Sam. What is Flash Memory? Flash memory is a type of memory storage device for computers & devices. It is a type of 'EEPROM' chip,
Course Overview This course encourages candidates to develop their knowledge and understanding of computer systems, the principles of computing (including.
W4118 Operating Systems Instructor: Junfeng Yang.
Computer Hardware. 11 th Class Computer Office Application-1 Computer Hardware Md. Jamirul Islam Assistant Professor Computer Monwara Azmat Ali College,
A computer consists of five functionally independent main parts.
Memory and Programmable Logic
Problem and Motivation
What is a computer? Simply put, a computer is a sophisticated electronic calculating machine that: Accepts input information, Processes the information.
基于多核加速计算平台的深度神经网络 分割与重训练技术
You can check broken videos in this slide here :
Machine Learning for dotNET Developer Bahrudin Hrnjica, MVP
Electronics Technology
Israel Cidon, Ran Ginosar and Avinoam Kolodny
RAID Redundant Array of Inexpensive (Independent) Disks
ASIC² Research Group Project: Design of peripheral circuit of memory array to perform in-memory logic operations Background: A memristive device (or a.
Presentation transcript:

In-memory Accelerators with Memristors Yuval Cassuto Koby Crammer Avinoam Kolodny Technion – EE ICRI-CI Retreat May 8, 2013 PU MEM NVM

3-way Collaboration A. Kolodny Y. Cassuto K. Crammer ML App. Devices Representations

The Data Deluge Mobile, Cloud Computing

Non-Volatile Memories 101 functionality density PROMEPROM E 2 PROM Memristors Mass Storage NAND Flash + logic!

Non-Volatile Memories 101 functionality density PROMEPROM E 2 PROM NAND Flash Main Memory Memristors + logic!

Memristor Crossbar Arrays

VgVg RLRL VoVo c ij c ij =0  high resistance  low current sensed c ij =1  low resistance  high current sensed Memristor Readout

VgVg RLRL VoVo Desired Path Sneak Path 1 1 c ij =0  high resistance  low current sensed c ij =1  low resistance  high current sensed Sneak Paths

Two Solutions Poor capacity High read power

Our Mixed Solution YC, E. Yaakobi, S. Kvatinsky, ISIT 2013 b

Results Summary YC, E. Yaakobi, S. Kvatinsky, ISIT ) Fixed partition2) Sliding window Higher capacity e.g vs for b=7 Column-by-column encoding, optimal

In-memory Acceleration Motivation: transfer bottlenecks Method: compute in memory, transfer results What to compute?

Similarity Inner Products Hyp. 1 Hyp. 2 Trial ∑ = ∑ =5 More similar Less similar

Inner Products in ML

Memristor Inner Products (ideal) Trial Hyp R= ∞ G T =3/2R R 2R Output = 3· ConstInner product

Ideal Inner Products Hamming distance in 3 measurements : 1 2 3

Real Inner Products Error terms

Evaluation Can compute Hamming distance as if ideal –3 measurements –plus arithmetic Cannot compute inner product precisely in 1 measurement

Continued Research Transform input vectors to maximize precision ML Theory: provable optimality (information-theoretic learning) ML Practice: optimize transformations within real ML algorithms

Multi-level Inner Products R= ∞ R1R1 R1+R2R1+R2 R2R2 R3R3 R3+R1R3+R1 2R 3

Thank You!