New Directions in Memory Architecture June 12, 2014



Similar documents
Accelerating I/O- Intensive Applications in IT Infrastructure with Innodisk FlexiArray Flash Appliance. Alex Ho, Product Manager Innodisk Corporation

Big Data Performance Growth on the Rise

ebay Storage, From Good to Great

Scaling from Datacenter to Client

Samsung 2bit 3D V-NAND technology

Redefining Flash Storage Solution

Enabling the Flash-Transformed Data Center

Flash s Role in Big Data, Past Present, and Future OBJECTIVE ANALYSIS. Jim Handy

NAND Flash Architecture and Specification Trends

Flash Technology Update from Micron and Intel

How SSD Can Improve Data Center Efficiency. Jim Elliott Samsung Semiconductor, Inc Vice President Memory Marketing

Module 2. Embedded Processors and Memory. Version 2 EE IIT, Kharagpur 1

Comparison of NAND Flash Technologies Used in Solid- State Storage

Samsung 3bit 3D V-NAND technology

Technology Trends in the Storage Universe

Operating Systems: Basic Concepts and History

The search engine you can see. Connects people to information and services

Storage Sizing Issue of VDI System

Boost Database Performance with the Cisco UCS Storage Accelerator

Journey to the All-Flash Data Center

Solid State Technology What s New?

Flash 101. Violin Memory Switzerland. Violin Memory Inc. Proprietary 1

Computer Graphics Hardware An Overview

3D NAND Technology Implications to Enterprise Storage Applications

NAND Basics Understanding the Technology Behind Your SSD

Operating Systems. Cloud Computing and Data Centers

Why Hybrid Storage Strategies Give the Best Bang for the Buck

Overview: X5 Generation Database Machines

Performance Beyond PCI Express: Moving Storage to The Memory Bus A Technical Whitepaper

Main Memory & Backing Store. Main memory backing storage devices

High-Performance Nested Virtualization With Hitachi Logical Partitioning Feature

Cloud Data Center Acceleration 2015

Enabling the Use of Data

SOLUTION BRIEF AUGUST All-Flash Server-Side Storage for Oracle Real Application Clusters (RAC) on Oracle Linux

How To Write On A Flash Memory Flash Memory (Mlc) On A Solid State Drive (Samsung)

Memory Channel Storage ( M C S ) Demystified. Jerome McFarland

CPS104 Computer Organization and Programming Lecture 18: Input-Output. Robert Wagner

Current Status of FEFS for the K computer

MaxDeploy Ready. Hyper- Converged Virtualization Solution. With SanDisk Fusion iomemory products

enabling Ultra-High Bandwidth Scalable SSDs with HLnand

Flash Memory Jan Genoe KHLim Universitaire Campus, Gebouw B 3590 Diepenbeek Belgium

The Flash-Transformed Financial Data Center. Jean S. Bozman Enterprise Solutions Manager, Enterprise Storage Solutions Corporation August 6, 2014

Sun Constellation System: The Open Petascale Computing Architecture

Data Center Solutions

Computer Systems Structure Main Memory Organization

Seagate Kinetic Open Storage Platform. James Hughes and many others

Managing the evolution of Flash : beyond memory to storage

1. Memory technology & Hierarchy

Amazon EC2 Product Details Page 1 of 5

Can Flash help you ride the Big Data Wave? Steve Fingerhut Vice President, Marketing Enterprise Storage Solutions Corporation

A New Chapter for System Designs Using NAND Flash Memory

Chapter Introduction. Storage and Other I/O Topics. p. 570( 頁 585) Fig I/O devices can be characterized by. I/O bus connections

How To Get A Memory Memory Device From A Flash Flash To A Memory Card (Iomemory) For A Microsoft Flash Memory Card From A Microsable Memory Card For A Flash (Ios) For An Iomemories Memory

September 25, Maya Gokhale Georgia Institute of Technology

WEBTECH EDUCATIONAL SERIES

Accelerating Real Time Big Data Applications. PRESENTATION TITLE GOES HERE Bob Hansen

FLASH TECHNOLOGY DRAM/EPROM. Flash Year Source: Intel/ICE, "Memory 1996"

Accelerating Microsoft Exchange Servers with I/O Caching

Benefits of Solid-State Storage

The Shortcut Guide to Balancing Storage Costs and Performance with Hybrid Storage

HP Z Turbo Drive PCIe SSD

Data Center Solutions

Cloud-Based Apps Drive the Need for Frequency-Flexible Clock Generators in Converged Data Center Networks

A Close Look at PCI Express SSDs. Shirish Jamthe Director of System Engineering Virident Systems, Inc. August 2011

OBJECTIVE ANALYSIS WHITE PAPER MATCH FLASH. TO THE PROCESSOR Why Multithreading Requires Parallelized Flash ATCHING

Price/performance Modern Memory Hierarchy

The Evolving NAND Flash Business Model for SSD. Steffen Hellmold VP BD, SandForce

WHITE PAPER 1

Flash & DRAM Si Scaling Challenges, Emerging Non-Volatile Memory Technology Enablement - Implications to Enterprise Storage and Server Compute systems

Diablo and VMware TM powering SQL Server TM in Virtual SAN TM. A Diablo Technologies Whitepaper. May 2015

io3 Enterprise Mainstream Flash Adapters Product Guide

Intel Solid-State Drive 320 Series

HP 80 GB, 128 GB, and 160 GB Solid State Drives for HP Business Desktop PCs Overview. HP 128 GB Solid State Drive (SSD)

How SSDs Fit in Different Data Center Applications

An Analysis Of Flash And HDD Technology Trends

A Look Inside Smartphone and Tablets

Where IT perceptions are reality. Industry Brief. Ride the Next Big Wave of Storage Networking. Featured Technology

Winbond W971GG6JB-25 1 Gbit DDR2 SDRAM 65 nm CMOS DRAM Process

Intel SSDs Transforming Storage in the Data Center

Cisco UCS B460 M4 Blade Server

Video Surveillance Storage and Verint Nextiva NetApp Video Surveillance Storage Solution

The Flash Transformed Data Center & the Unlimited Future of Flash John Scaramuzzo Sr. Vice President & General Manager, Enterprise Storage Solutions

THE DEFINITIVE GUIDE FOR AWS CLOUD EC2 FAMILIES

Memory Basics. SRAM/DRAM Basics

Driving Big Data with OCZ Enterprise SSDs

SAP HANA - Main Memory Technology: A Challenge for Development of Business Applications. Jürgen Primsch, SAP AG July 2011

and the Dependability Implications

2 CMOS 5 V/5 V, SPST Switches ADG601/ADG602

Communicating with devices

PCI Express Basic Info *This info also applies to Laptops

Data Center Storage Solutions

SSDs tend to be more rugged than hard drives with respect to shock and vibration because SSDs have no moving parts.

Contents. 1. Trends 2. Markets 3. Requirements 4. Solutions

EMC VMAX3 FAMILY - VMAX 100K, 200K, 400K

EMC FLASH STRATEGY. Flash Everywhere - XtremIO. Massimo Marchetti. Channel Business Units Specialty Sales EMC massimo.marchetti@emc.

Emerging storage and HPC technologies to accelerate big data analytics Jerome Gaysse JG Consulting

ECLIPSE Performance Benchmarks and Profiling. January 2009

An Introduction to High-Frequency Circuits and Signal Integrity

Riding silicon trends into our future

Advances in Flash Memory Technology & System Architecture to Achieve Savings in Data Center Power and TCO

Transcription:

New irections in Memory rchitecture June 12, 2014 Bob Brennan, enior Vice President Memory olutions ab Bob.Brennan@amsung.com

egal isclaimer This presentation is intended to provide information concerning memory industry trends. We do our best to make sure that information presented is accurate and fully up-to-date. owever, the presentation may be subject to technical inaccuracies, information that is not up-to-date or typographical errors. s a consequence, amsung does not in any way guarantee the accuracy or completeness of information provided on this presentation. amsung reserves the right to make improvements, corrections and/or changes to this presentation at any time. The information in this presentation or accompanying oral statements may include forward-looking statements. These forward-looking statements include all matters that are not historical facts, statements regarding the amsung Electronics' intentions, beliefs or current expectations concerning, among other things, market prospects, growth, strategies, and the industry in which amsung operates. By their nature, forward-looking statements involve risks and uncertainties, because they relate to events and depend on circumstances that may or may not occur in the future. amsung cautions you that forward looking statements are not guarantees of future performance and that the actual developments of amsung, the market, or industry in which amsung operates may differ materially from those made or suggested by the forward-looking statements contained in this presentation or in the accompanying oral statements. In addition, even if the information contained herein or the oral statements are shown to be accurate, those developments may not be indicative developments in future periods. amsung 2

genda» Environment BW & Capacity growth» M BW & Capacity -> Tiering» lash cales, Gets Intelligent, Tiers» New Persistent Performance amsung 3

Environment: Mobile & Cloud Information growth drivers over time Internet of everything! 2012: Mobile connected devices exceeded the world's population amsung 4

Environment: atacenter Infrastructure More applications for data ata traffic: 78% CG * 1000 PB: 1EB (10 18 ) Billions of evices! More video is uploaded to YouTube in one month than the 3 major U networks created in 60 years ource: Cisco Visual Networking Index What about Exabytes? 5 EB: Total data created between the dawn of civilization and 2003 amsung 5

EXBYTE VMs per ost Environment: Escalating emand for M and torage In-Memory nalytics for Big ata Growing x86 erver Virtualization ensity orecast % of Installed Workloads unning VM ensity in a VM per ost erver Unstructured ata tructured ata ource: EMC and IC ource: Gartner and IC Escalating Memory-Intensive Workloads ata Center Processor Growth PC Graphics 2x Volume Growth inancial Gaming Big ata ource: Intel amsung 6

Environment Bandwidth emand isplay Camera Video N-screen (1920x1080) 13MP 1080p -... U (3840x2160) 20+MP 4K U Memory Bandwidth equirements Now 7.5x (~100GB/s) 2018 100x (~1.4TB/s) U TV 400~600 Mbps 10~20GB/s 3.7x (~1.6Gbps) 12.5x (~5.3Gbps) Peta-flops 20Peta-flops Exa-flops GB/s/node Gb/s/pin [ource: Memory systems for Petalop to Exalop class machines by IBM, 2007 & 2010] Mobile: isplay/gx/camera Exponential Bandwidth emand erver: Core caling inear to Exponential Bandwidth emand amsung 7

Environment Capacity emand 1GB 2GB 3GB 4GB 100~200TB 2~4GB Memory Capacity equirements Peta-flops 20Peta-flops Exa-flops Memory Capacity/ystem Now >5x (~750TB) >4x (~16GB) 2018 >70x (~10PB) >32x (~128GB) Memory Capacity/Node [ource: Memory systems for Petalop to Exalop class machines by IBM, 2007 & 2010] Mobile: isplay/gx/camera ~inear Capacity emand erver: Core caling inear - Exponential Capacity emand amsung 8

genda» Environment BW & Capacity growth» M BW & Capacity -> Tiering» lash cales, Gets Intelligent, Tiers» New Persistent Performance amsung 9

The Trade-off Triangles Bandwidth IOPs Power Power atency Capacity Endurance Capacity M Non-Volatile amsung 10

Bandwi dth Power M: Bandwidth caling atenc y Capacit y [Mbps] 4200 ubject to cost/energy efficiency, scaling, 3600 Optical (?) 3200 2400/2667 2133 1866 1600 1333 4 5 (?) & New I/ (?) Wall? Multi-rop Bus Challenge: igher BW, ower V 2011 2015 2018 [Year] New olution Needed amsung 11

Cell T leakage Current Bandwi dth Power M: caling Challenges atenc y Capacit y W/ B/ 1 Plate efresh ifficult to build high-aspect ratio cell capacitors decreasing cell capacitance eakage current of cell access transistors increasing tw Contact resistance between the cell capacitor and access transistor increasing On-current of the cell access transistor decreasing Bit-line resistance increasing VT s cell capacitance shrinks, more frequent Time amsung Innovation needed to scale to 10nm & beyond 12

Bandwi dth Power M: atency Challenge atenc y Capacit y [tc, ns] 60 ubject to cost/energy efficiency, scaling, 50 40 30 20 10 2005 2010 2015 2020 ~Constant ow atency Needed [Year] amsung isruptive olution Needed 13

M: Go Wide for Bandwidth ITEM Mobile WIO2 BM (igh B/W Memory) M Base die + M WIO2 P i Interposer GPU PKG PCB BM Base Pin #Bump Bottom die N/ Buffering & ignal re-routing BW (GB/s) 25.6~51.2 128~256 peed 0.4~0.8 Gbps 1~2 Gbps # I/O 512 1,024 ogic 1~2K 6K~8K M 1~2K ~3K Cube (GB) 1 / 2 1 / 2 / 4 # TV stack 1 / 2 / 4 1 / 2 / 4 M density 8Gb 8Gb pplica tion GX card UT - PC - erver - (Cache) Mobile - Bandwi dth Power amsung atenc y Good BW & atency till Need Capacity Capacit y 14

M: ybrid Memory ystems Mobile erver OC CPU Tiered Memory Controller atency Bandwidth Power Capacity CPU CPU CPU CPUs CPU CPU Tiered Memory Controller WIO? WIO? WIO? igh BW M igh Bandwidth Tier 4 B M? P 4 P 4 P 4 P 4 igh Capacity Tier C M C M amsung Tiered Capacity, Tiered atency, T-M? 15

1 st tep: ystem Tiering M 16 igh Performance Tier igh Capacity Tier M M M amsung BM

genda» Environment BW & Capacity growth» M BW & Capacity -> Tiering» lash cales, Becomes Intelligent, Tiers» New Persistent Performance amsung 17

lash IOPs Power Endurance Capacity amsung 18

IOPs Power lash: Capacity caling Endur ance Capac ity 120nm 1Gb 90nm 2Gb 70nm 4Gb 60nm 8Gb 50nm 16Gb Cost of Patterning 40nm 32Gb 19nm 128Gb amsung caling Becomes ifficult Need a New olution 19

ow ec. Breakthrough: 128Gb V-NN - Vertical-NN Technology - Chip ize : 133mm 2 0.96Gb/mm 2 64Gb rray Plane-0 64Gb rray Plane-1-24-W tacked ayers - 64Gb rray 2-Plane - One-sided Page Buffer : (8KB x 2) Page ize Page Buffer Peripheral Circuits Page Buffer - synchronous Interface : Wave-pipeline datapath : 667Mbps at Mono ie : 533Mbps at 8-stacked ies amsung World s 1 st 3 V-NN Mass Production lash 20

V-NN rray tructure dvanced V-NN Technology with amascened Metal Gate Cell : ll-around Gate tructure + Charge Trap lash tring : 24-W + 2-W + 2-elect W Block : 8 trings with hared B (8KB) B<p> B<q> <7> <1> <0> ummy1 W23 Poly channel CT dielectrics W0 ummy0 Metal gate G C amsung 21

V-NN eatures Bits per Cell 2 ensity Technology 128Gb Three imensional Vertical NN, 3-metals Organization 8KB 384 pages 5464 blocks 8 Program Performance ata Interface peed Power upply 50MB/s for Embedded pp., 36MB/s for Enterprise 667Mbps@Mono, 533Mbps@8-stack Vcc=3.3V / Vccq=1.8V amsung

Conventional Ext.V Normalized (a.u.) Measured ctive Power Improves Over 50% ower Energy dvantage is achieved Increasing overall Performance by using 8-way Interleaving NN Operation 43% 54% 53% ICC1 (ead) ICC2 (Program) ICC3 (Erase) amsung 23

Normalized (a.u) Enterprise Comparison 512GB Ep- maller eal Estate Planar NN (8-ch, 8-way) 3 V-NN (8-ch, 4-way) 22% aster 20% aster 27% ower 45% ower igher Performance equential Write (MB/s) andom Write (IOP) verage Power (mw) Peak Power (mw) amsung 24

lash: caling Continues esign ule (nm) 2 Planar 3 V-NN / No Patterning imitation 8 stack 24 stack 16Gb 128Gb 1Tb 128Gb 03 05 07 09 11 13 15 17 Year amsung Capacity, Endurance, Power 25

# of Cells Normalized (a.u.) Endur ance IOPs Power Capac ity NN lash Endurance lash: MC Endurance 36MB/s + 35K Endurance for ata-center & Enterprise pplications 50MB/s + 3K Endurance for Mobile pplications Planar 1xnm NN after 3K cycle vg. tpog=0.33ms (50MB/s) vg. tpog=0.45ms (36MB/s) 3 V-NN after 35K cycle Vth (a.u) Time [us] amsung Endurance improved dramatically 26

IOPs Power lash: Performance Endur ance Capac ity atency & IOP Interface & Performance ms 16 14 12 10 8 6 4 2 0 >100x <0.3ms 7.2K PM 15K PM 12000 10000 8000 6000 4000 2000 0 MB/s 5000 4500 4000 3500 3000 2500 2000 1500 1000 500 0 2004 2006 2008 2010 2012 2014 2016 PCIe x4 T 2005 2008 2011 2014 otational atency VG eek IOP Interface Unlocks Bandwidth: PCIeG2->G3->G4 olution needs to scale: Controllers, lgorithms, & lash Organization amsung Increasing Intelligence & ophistication 27

lash: Inherent Intelligence Intelligent IOPs IOPs Power Endurance Endurance Capacity 3 caling amsung 28

2 nd tep: ystem Tiering lash/s 29 igh Performance Tier igh Capacity Tier M M M Intelligent lash Tier amsung

Today s ack caling cknowledgement: Krishna Malladi. isclaimer: conceptual model only. CPU data on different scale. amsung lash ignificantly Improves the M-isk Gap 30

genda» Environment BW & Capacity growth» M BW & Capacity -> Tiering» lash cales, Becomes Intelligent, Tiers» New Persistent Performance amsung 31

atency (ns) Opportunity for New Technology 10,000,000 1,000,000 lash 100,000 10,000 1,000 Persistent Performance 100 M 10 C 1 20 40 60 Bandwidth (GB/s) amsung 32

TT-MM etention tandby Power M TT-MM peed (/WT) Outer Better Tech. Maturity TT-MM Cell tructure Bit cost (calability) Endurance Promising Technology, Not Mature Yet amsung 33

Persistent Tiered Caching 3 rd tep: New possibilities 34 igh Performance Tier igh Capacity Tier M M M Intelligent lash Tier N E W N E W N E W N E W N E W N E W N E W N E W Persistent Performance, Byte addressable Intelligent lash Tier (PCIe) igher Performance Tier M M M N E W N E W N E W amsung Intelligent lash Tier ()

uture ack caling Vision cknowledgement: Krishna Malladi. isclaimer: conceptual model only. amsung Ideal caling: 1. V-NN 2. NMT 3. ystem W 35

Thank you! Questions: Bob.Brennan@amsung.com amsung 36