To create this report, the authors performed thorough and exhaustive research into the technology and data leading up to its introduction, including reviews of technical conference presentations, patents, and trade shows, along with in-depth interviews with participants in the established and emerging memory markets.
The market was modeled to produce forecasts by estimating its demand in servers, which the companies say is their key market and analyzing which parts of the DRAM market will be offset with this new memory type, and comparing that to the lost market for server DRAM.
Key findings are:
1. The market for 3D XPoint memories could optimistically grow as large as $2.12 billion in 2019, only 2 years after its first production-volume shipments.
2. This technology will not grow unless prices fall below those of DRAM, and this will require significant resources and volume.
3. The Server DRAM market would suffer the greatest losses to this technology, with 2019 revenues hitting their lowest level since 2012.
4. Other markets, namely SSD, HDD, and processors, will not be noticeably impacted by the use of 3D XPoint memory.
5. Despite the fact that most emerging memories are being positioned as NAND flash replacements, Intel and Micron are positioning 3D XPoint as a new layer in the memory/storage hierarchy.
6. Little of the product should ship before 2017, the year the report calls for a memory price collapse.
7. Because the product’s production ramp coincides with this collapse, 3D XPoint will have little negative influence on DRAM pricing.
8. Significant system and software support is needed to make this a market reality.
2. What Is 3D XPoint Memory?
- What is a Crosspoint Memory?
- The “Sneak Path” and Selectors
- Why Are Selectors So Challenging?
- On/Off Current
- I/V Characteristics
- Process Technologies
- Stacking Crosspoint Decks
- What is the Memory Cell Made Of?
3. A Brief History of PCM
- How Will It Be Used?
- Improving the Memory/Storage Hierarchy
- A Premature Revelation
- How Can Optane be DDR4 Compatible?
- A Strong Focus on Gaming PCs
- How does Optane Impact Performance?
- NVMe XPoint Matches Clients’ Needs
- Important Write IOPS Cost Savings
- How Important is Persistence?
- Nonvolatility for Power Fail
- Nonvolatility for Data Resilience
- Persistence in I/O-Bound Systems
4. Current State of Nonvolatile Memory Support
- In Summary: The Market Doesn’t Yet Exist
- How Will The Market Develop?
- Why is Intel Getting Back Into Memories?
- How Real is 3D XPoint?
- What Systems Will Use it?
- The Issue of Sole-Sourcing
- Impact on Other Markets
- NAND Flash and HDD
- Cost Adder
- Achieving Sub-DRAM Costs
- The “Chicken & Egg” Problem
5. The Intel/Micron Relationship
- Is Micron’s Prospective Acquisition An Issue?
- Getting Ahead of Samsung
- HP’s “The Machine”
- What is a Memristor?
- Memristor Basics
- Implications for the 3D XPoint Memory
- What This Means to OEMs
- What This Means to Computer Purchasers
- What This Means To The Memory Business
- Further Reading
List of Tables
Table 1. HP's Calculation of Software Latency Penalty
Table 2. Server DRAM Without 3D XPoint
Table 3. Server DRAM With 3D XPoint
Table 4. Server DRAM Market Impact of 3D XPoint
List of Figures
Figure 1. 3D XPoint Internal Diagram
Figure 2. Electron microphotograph of 3D XPoint memory
Figure 3. Reading a bit in a crosspoint array
Figure 4. "Sneak Path" current causes erroneous bit read
Figure 5. More Detailed Depiction of 3D XPoint Memory
Figure 6. I/V curve of a bipolar selector for resistive RAMs
Figure 7. Multiple Patterning for Finer Process Geometries
Figure 8. Mock-up of 3D XPoint memory at the Intel Developer Forum
Figure 9. The Intel/Numonyx PMCS, IEDM 2009
Figure 10. Periodic Table With All Possible Candidates for Memory Production
Figure 11. PCM Cover Article, Electronics Magazine, September 28, 1970
Figure 12. Intel's Optane offerings and their placement in the memory/storage hierarchy
Figure 13. Leaked Purley Memory Information
Figure 14. Drawing from Intel Patent Application 20,140,304,475
Figure 15. Latencies of Various Storage Types
Figure 16. Carving out a space for 3D XPoint memory
Figure 17. Throughput vs Queue Depth of NAND and Optane NVMe SSDs
Figure 18. The Cost of a Write Throughput in IOPS
Figure 19. The Memory/Storage Hierarchy of Computing Applications
Figure 20. Historical Average Price per Gigabyte of NAND and DRAM
Figure 21. Ratio of DRAM to NAND Price per Gigabyte
Figure 22. NAND Filled the Gap between HDD and DRAM
Figure 23. Micron's view of the memory/storage hierarchy with 3D XPoint memory . 44
Figure 24. 3D XPoint is intended to fill a similar gap between NAND flash (SSD) and DRAM
Figure 25. Micron August 18, 2015, Summer Analyst Day Slide
Figure 26. Server DRAM Gigabyte Shipments With and Without 3D XPoint
Figure 27. Server DRAM Revenues With and Without 3D XPoint
Figure 28. Server 3D XPoint Revenue Forecast (Aggressive Scenario)
Figure 29. Micron's DRAM yield ramp curves
Figure 30. DRAM Learning Curve: The more you make, the cheaper it gets
Figure 31. Hwang's Law slide from 2005
Figure 32. Relationship between Memristors, Resistors, Capacitors, and Inductors
Figure 33. Two ways to achieve the same performance DRAM-rich and Flash-rich
Figure 34. Different performance levels of two similarly-priced systems
- Conference ConCepts
- Cypress Semiconductor
- Flash Memory Summit