Emerging Non-volatile Memories: Opportunities and Challenges Shahar Kvatinsky Department of Electrical Engineering Technion – Israel Institute of Technology November 11, 2015

The Problem with NAND Flash Source: D. Ruggeri, MICRON Flash memory 1x 3D NAND today. Going back in technology. http://www.eetimes.com/document.asp?doc_id=1280886 https://www.google.co.il/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&ved=0CCsQFjAA&url=https%3A%2F%2Fwww.micron.com%2F~%2Fmedia%2FDocuments%2FProducts%2FWhite%2520Paper%2Fscaling_nonvolatile_mem_in_embedded_systems.pdf&ei=CL5jUuSMEcbQhAfZkoHgAQ&usg=AFQjCNGpTnUbnJyjYRF5aplC9-fYasKfFQ&sig2=m6MHtdCsUU7Pgh7ZePF7Lw&bvm=bv.54934254,d.ZG4&cad=rja http://www.memcon.com/pdfs/proceedings2013/keynotes/New_Directions_in_Memory_Architecture.pdf http://www.semiwiki.com/forum/content/2103-can-%93less-than-moore%94-fdsoi-provides-better-roi-mobile-ic.html

Going 3D Company 2012 2013 2014 2015 2016 21nm 19nm 16-19nm V-NAND (24 layers) 12nm V-NAND (32L) 12nm V-NAND (48L) 20nm 16nm 3D NAND 3D NAND Gen. 2 15nm 3D NAND BiCS http://www.3dincites.com/2013/08/getting-back-to-basics-a-look-at-3d-nand-and-3d-dram/ Toshiba’s BiCS (Bit Cost Scalable) 3D NAND "Technology Roadmap for NAND Flash Memory". techinsights. April 2014. Retrieved 9 January 2015.

Motivation for New Technologies Scalability of 3D Volatile technologies: Energy (leakage, refresh in DRAM) Limited density and scalability Chip energy http://www.3dincites.com/2013/08/getting-back-to-basics-a-look-at-3d-nand-and-3d-dram/ Toshiba’s BiCS (Bit Cost Scalable) 3D NAND "Technology Roadmap for NAND Flash Memory". techinsights. April 2014. Retrieved 9 January 2015. Source: E. Esteve, IPNEST

Emerging Nonvolatile Memory Technologies RRAM PCM STT MRAM CBRAM There is a need for new memory technologies – large interest – we will call all of them as memristors.

Resistive RAM (RRAM, ReRAM) Lim, E.W.; Ismail, R. Conduction Mechanism of Valence Change Resistive Switching Memory: A Survey. Electronics 2015, 4, 586-613. http://www.digitaltrends.com/computing/resistive-ram-how-it-could-change-storage-forever/

RRAM Status Product: Prototypes: 24nm 2-layer 32Gb (ISSCC 2013) 40us read, 230us write Research/development: There is a need for new memory technologies – large interest – we will call all of them as memristors.

Conductive Bridging RAM (CBRAM, PMC) Nanotechnology and Nanomaterials » "New Progress on Graphene Research", book edited by Jian Ru Gong, chapter 7 Low voltage/power

CBRAM Status Products: Prototypes: 27nm 16Gb (ISSCC 2014) 2us read, 10us write, 1T1R, 6F2 Research/development: There is a need for new memory technologies – large interest – we will call all of them as memristors.

Phase Change Memory (PCM, PCRAM, PRAM) There is a need for new memory technologies – large interest – we will call all of them as memristors.

PCM Status Products: Prototypes: Research/development: 20nm, 8 Gb, 2012 45nm, 1 Gb, 2012-14 There is a need for new memory technologies – large interest – we will call all of them as memristors.

Spin-Torque Transfer Magnetoresistance RAM (STT MRAM) http://www.azom.com/article.aspx?ArticleID=10010

STT MRAM Status Products: Prototypes: Research/development: 64Mb, DDR3, 400 MHz 1Mb, 45nm 32Mb There is a need for new memory technologies – large interest – we will call all of them as memristors.

Shared Properties BEOL CMOS compatibility Resistive Nonvolatility High endurance Speed Density Low voltage Retention There is a need for new memory technologies – large interest – we will call all of them as memristors.

Challenges Selectors Device variations Yield Power Endurance There is a need for new memory technologies – large interest – we will call all of them as memristors. Source: Crossbar, IEDM 2014

Immediate Applications NAND Flash replacement Storage Class Memory Embedded NVM There is a need for new memory technologies – large interest – we will call all of them as memristors.

On-Die Intensive Memory Circuits and Architectures Multistate Register Continuous Flow Multithreading

FPGA Source: P.-E. Gaillardon, EPFL There is a need for new memory technologies – large interest – we will call all of them as memristors. Source: P.-E. Gaillardon, EPFL

Integrating memristors with standard logic Logic with Memristors Memristor layer CMOS layer Integrating memristors with standard logic CMOS logic is good, so why bother with memristor logic? First reason – if the memristor-based memory is there, we can do logic in the same price! We can break the conventional Von Neuman architecture to new architecture for calculation inside the memory. It can be done in parallel and in a dynamic manner. Second reason – we can use memristor-based logic as a redundant logic layer for backup or fixing soft errors for example and to save die area – ECC, Die area, etc. And the last reason – it is fun and interesting, after all I am in the academia… Logic within memory Beyond Moore Beyond Von-Neumann

Neuromorphic Computing This slide is for short presentations

Circuits Source: T. Prodromakis, University of Southampton There is a need for new memory technologies – large interest – we will call all of them as memristors. Source: T. Prodromakis, University of Southampton

Summary Need for new nonvolatile memory technology Challenges Opportunities for novel applications: Logic Programmable circuits Neuromorphic On-die memory There is a need for new memory technologies – large interest – we will call all of them as memristors.

Thanks! shahar@ee.Technion.ac.il

Cost Random Access Time ns us ms s CPU Register CPU L1-3 Cache DRAM SCM SSD Cost HDD Diameter ~log capacity ns us ms s Random Access Time