Memory & Storage Challenges and Solutions
G S A 2 0 1 9
Jinman Han
Senior Vice President, Memory Product Planning & Application Engineering Legal Disclaimer
This presentation is intended to provide information concerning SSD and memory industry. We do our best to make sure that information presented is accurate and fully up-to-date. However, the presentation may be subject to technical inaccuracies, information that is not up-to-date or typographical errors. As a consequence, Samsung does not in any way guarantee the accuracy or completeness of information provided on this presentation.
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 Samsung Electronics' intentions, beliefs or current expectations concerning, among other things, market prospects, growth, strategies, and the industry in which Samsung 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. Samsung cautions you that forward looking statements are not guarantees of future performance and that the actual developments of Samsung, the market, or industry in which Samsung 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. Abstract
Memory-centric system innovation is the overarching theme of modern semiconductor technology and is one of the crucial driving forces of the future IT world. As one of the foundational elements of the future IT industry, modern memory and storage technology remains critically important.
Various new technologies based on DRAM and NAND have been developed for future application, while the limitation of the trade-off between performance and cost metrics has driven attention to the search for optimal solution.
Therefore, A deep understanding of the memory industry and memory technology is crucial for accelerating the new data centric world. In this talk, current memory and storage challenges and solution are addressed. Various cutting-edge memory and storage technology and their implications are also investigated. General Trend
Industry Change Memory Change Big Changes on Industry
Smart Factory Smart Auto Bigger Home data Process time & energy ↑ Memory
Processor-centric Memory-centric Big Changes on Memory
One size can’t fit all Memory Hierarchy change 1) One size can’t fit all (# of applications)
Auto Smart Factory
Health Smart
Home
………...……..
..
…...
…..… ………… Various Memory Requirements
5G Reliability Thermal AI
High Performance Thin Package Low Cost
TB
GB High Capacity Low Power Auto Datacenter Diversification on DRAM
HBM HBM2
LPDDR5
LPDDR GDDR6 GDDR DDR
High Capacity DDR4 DIMM
PC Era Mobile Era Datacenter Era AI Era DRAM Solutions
HBM2 GDDR6
8H stack HBM2 over 1TB/s system memory bandwidth 18Gbps Highest speed Compact & Power efficient package 864GB/s system memory bandwidth
LPDDR5 DDR4
over 50GB/s bandwidth 256GB max density 40% lower power (compare with LPDDR4X) Align with High capacity requirement Diversification on Storage
SATA SSD
PC Era Datacenter Era AI Era Storage Solutions
Z-SSD eUFS
Z-NAND based solution 24Gbps Performance <100us latency in storage system Low Power consumption for mobile
QLC SSD 3D NAND
High Capacity solution Outstanding Performance as Read-Intensive warm data storage with 120+ vertical stacking
2) Memory Hierarchy Change GB]
Cost [$/
Performance Cost Cost
DRAM
SCM “Attack of the Killer ㎲” NAND Luiz Barroso et.al.
HDD
SCM: Storage Class Memory ㎲ Performance New Memory Solutions as Gap Filler
Traditional Hierarchy New Hierachy
Small # of Huge # of SW based
Software
Hardware
Technology Collaboration (/each component) (in system level) New Memory Solution : Z-SSD
PERFORMANCE Z-SSD™ Read Latency Read QoS–99.999% For Data Analytics & Artificial Intelligence (4KB Random, FIO) (Mixed Random)
Samsung Z-NAND 5.5x 100x Lower Lower
New CONTROLLER
Fast High < 15 us Response Bandwidth < 300us
TLC based Z-SSDTM TLC based Z-SSDTM
Measured by Samsung Many Challenges Ahead Performance Power Reliability
Blocksize Customized Form Factor Solution DRAM Challenges
Performance Power Reliability Ever-Increasing Memory Performance
700 Require more memory B/W 600 - Number of channel ↑ System memory - Per pin speed ↑ requirement 500
400 BW gap
300 Performance (GB/s) Performance 200 Memory BW
100
0 `11 `15 `20 `25 (year) DRAM Capacity Increase within Power Budget
② Additional enablers - New technology
- Arch. optimization Power (w) Power
① Traditional enablers - Voltage down - Process scaling Limit by industry rack infrastructure for 1U/2U (15W)
64GB 128GB 256GB 512GB 768GB 64GB 128GB 256GB 512GB 768GB DDR3 DDR4 DDR4 DDR5 DDR5 (4GbxQDP) (8Gbx4H) (16Gbx4H) (16Gbx8H) (24Gbx8H) [email protected] [email protected] [email protected] [email protected] [email protected] Continuous Scaling with better RAS Solution
Cell Capacitance 1 O/H
with Redundancy cell Chip sizeO/H Chip
Most effective way to achieve DRAM reliability target
O/H with On-die ECC
0 30xnm 2xnm process 1xnm process and later NAND Challenges
Blocksize Power What will What will # of Stacked WLs happen ~100 2019 ? V - Stack Up Stack ~500 2025 Block Size Increase
1 Block How to manage a Huge Block? 400 ~10x
1 Block 200
1x Block Size 0 20192018 20252025+ Power Consumption Increase
How to reduce the impact of ~5x # of WLs? ~2x
1x
Power derived from # of WLs ↑
2019 2025 Storage Solution Challenges
Performance Customized Next Scalibility Solution Form Factor Scalability on Performance over Capacity
32ch CTRL ??
External Bandwidth = Internal Bandwidth
Industry
16ch requires Performance Scalability Hetero-Media High-Speed NAND SSD 8ch
NAND Capacity
16TB 32TB 64TB 128TB @’16 @’18 Platform Architecture for Customized Solutions
Workload Read-Intensive Read/Write Mixed Write-Intensive Pattern Sequential 100% Mixed Random 100% Applications DWPD 0.1 0.3 0.5 1 3 5 10 dependent
Power Energy-Oriented Performance-Oriented Requirements
Feature SR-IOV IOD ZNS Customized Features ∙∙∙
S/W Feat. Security Z-NAND MLC/TLC QLC Ch Module-1 Ch Module-2 Power Switch Buffer Next Form Factors of SSD
Performance Tier Caching Tier NF1
E1.S HHHL : Extendability limit HHHL
E1.L U.2 ? U.2 : Connector SI limit E3 Capacity Tier M.2 Booting M.2 : Power limit Open Standard
Semiconductor sales exceed DDR2 DDR3 DDR4 NVDIMM-P DDR5 $200b SATA PCIe NVMe PCIe PCIe $100b 1.0 2.0 1.1 5.0 $1b 4.0 PCIe SATA SATA PCIe NVMe NVMe NVMe 1.0 2.0 3.0 3.0 1.0 1.2 1.3 ‘66 ‘94 1999 ‘00 2003 2004 2005 2007 2008 2010 2012 2013 2014 2016 2017 2018 2019
UTRAM
RDRAM OneNAND OneDRAM PCM S/W Defined Turbo Hybrid Memory FlashDIMM DIMM Memory EDRAM Low Latency DRAM DDR-T Key Summary
• Diversifying workloads and requirements require multiple memory/storage solutions
• To bring about hierarchical changes, ecosystems must get ready
• If we think about 5 years ahead, technical challenges loom large
• Open standards are critical, but there’re too many de-facto standards “We will find a way, we always have…“
– Interstellar, 2014 -