Thursday, November 1, 2018

Intels Announcement of Beta Deployment of Optane Persistent Memory

Intel announce that Optane Persistent Memory is shipping to select customers and those companies will deploy solutions "soon".

What was announced:
Optane Persistent Memory is shipping as Beta Program to select customers.
Widespread shipping in 2019
More importantly, they announce two ways in which it will work.

1) Memory Mode: Big and Affordable, but Volatile
This is 100% lined up with what we forecast. Optane Memory main value proposition is adding tons of memory at slower speed in DIMM format on the memory bus. Half the price of DRAM, probably 7x higher latency, much higher density. 6TB is possible. You make the speed OK by using DRAM cache. Example:  1.5TB of Optane with 192GBs of DRAM. but it only appears as 1.5TBs of memory since the DRAM is all used as cache. The downside? The memory is not persistent. Since you use a volatile cache, you can lose the data. 

Summary: Optane Persistent Memory in Memory Mode is not persistent. 

2)  App Direct Mode: Big, Affordable, and Persistent

In this mode, you have Optane in addressable memory. you have DRAM as well each is addressable separately. If you want to write to 192GBs of DRAM fine... just like always. If you want to write the the Optane, fine... its persistent. But you have to explicitly decide where to write. 

Summary: Optane Persistent Memory in App direct mode is persistent. 

Actual value proposition will depend on a few items:
A) Do you need tons of memory but dont need persistence? 
B) Do you need persistence and have the application software to manage it
C) Are you OK that the persistent memory is about 7x slower but is 50% cheaper

We have more data on actual speed, cost, and models on the next Gen Optane. We also previously published the models for Revenue from Optane SSDs adn Optane persistent memory.

Final Thought: If memory mode was persistent, we would actually have a simple memory solution. And something tells me there will be limitations built into the drivers on writing optane in App direct mode since it does not have infinite endurance. I am thinking that those details will come out when we see the end products... we will let you know.

Mark Webb

Intel announcement

Explanation of Modes

Tuesday, October 2, 2018

MKW Ventures Consulting Reports and Presentations

Out recent reports and presentations from Flash Memory Summit are published here

Persistent Memory
3D Xpoint Optane
Emerging Memory

Also detailed presentation on out cost model which is available for DRAM, NAND, 3D XPoint, and Other memory technologies.

Future reports on details of China's Memory technologies and plans coming

Many reports listed below

Saturday, September 29, 2018

Xpoint Optane Performance and Revenue

At 2018 Flash Memory Summit, I presented a summary of 3D Xpoint status. It includes applications, a model for the chip performance, and revenue projections. Summary is attached here

Some of my estimates. A model for what 3D Xpoint chips could be….

  • 128Gbit Chip with >10% overprovisioning on the chip itself
  • Read Latency: ~125ns, Write Latency: “higher”  (why higher is a long story)
  • Endurance: ~200K cycles spec with management techniques. Note: cycling capability is at a certain fail rate which is managed by controller.
  • 2nd Gen will have 2x the density at about 30% lower cost. It will be available in 2020
  • It is a very fast, high endurance, byte addressable, NVM replacement (at much higher cost than NAND)
  • Not DRAM replacement. 

Model is based on Optane Memory electrical analysis, Intel announcement and PCM physics 

So how high could the sales be? Revenue model

While high performance, high price SSDs are great, the volumes and revenue from this market are too low to pay for the development of the technology and manufacturing costs. Memory costs are all about volume efficiency paying for expensive development and fab capital expenditures.

Optane DIMMS are the best opportunity for Intel to achieve scale needed to make 3D Xpoint successful. The DIMMS require Cascade lake to work so we expect yet another "launch"* in December to signal Optane shipping as Apache Pass with Cascade lake processor.

Detailed presentation below from Flash Memory Summit Page ... I have other presentations there on persistent memory, and a comparison of all new memories performance and cost

Side Note: After FMS, Intel announced that it is moving 3D XPoint development to Rio Rancho, New Mexico, a site I worked at for the better part of 20 years. There is a separate discussion on why this was done and the implications to Micron and Intel. For now its just great to have memory development 20 minutes from my home!

Flash Memory Summit page

*Launch definition for many companies is quite different from what you might expect. Intel is no exception!

Mark Webb


Monday, August 13, 2018

Five Thoughts from 2018 Flash Memory Summit

Five thoughts from 2018 Flash Memory Summit

Of course, I have more than five but…. It’s a start!

I made more than a few presentations at the 2018 Flash Memory Summit on Persistent Memory, New technologies and NAND future. Documents will be posted. I also had some takeaways after discussions with attendees and got to see some friends from my past.

  • Persistent Memory is starting to see more products, applications, and potential than the initial NVDIMM-N. People are using NVDIMM-N for log and journal files and items that cannot be lost. New versions that combine some DRAM and lots of NAND or 3D Xpoint will be the future. Real Products available in 2018
  • 3D Xpoint, ReRAM, and MRAM are available today. 3D XPoint is being pushed by Intel and as a result will sell more than all other “New NVMs” combined. This volume will help create markets for all other NVMs. This is a huge change in the past 3-5 years.
  • New NVMs will struggle to ramp as fast as 3D XPoint as capital and expense spending is required to get these to high volume. Intel has spent ~$1B on getting 3D XPoint to sell. No other company can afford that. We need to find solutions and business models to help in this area. I presented a Product Lifecycle model for new memories that can be tied to investment and partnership plans.
  • Systems solutions, Fabric, SDS are allowing both faster and more economical use of persistent memory and Flash solutions in general. These will determine overall datacenter volume ramps in the future
  • YMTC presented their NAND technology and Fab Plans. Their challenge was already interesting to get into a very competitive market. The technology changes they are planning will add to the challenge and opportunities. We have a detailed report on technologies and the revenue we will see from YMTC over the next 5 years.

Some Highlights

Thursday, July 26, 2018

Xtacking from YMTC, Fast NAND and More

"Yangtze Memory Technologies to Debut New 3D NAND Architecture"

YMTC will announce their Xtacking technology at Flash Memory Summit Aug 7th per a company announcement. We will dig up details for you but it looks to be another version of "Fast NAND".

Fast NAND is what I refer to when we discuss Z-NAND (Samsung) or similar products being proposed from Toshiba, Hynix. These are aimed to compete with Intel Optane SSDs. Versions can also be used for DIMMS in a NVDIMM-N or NVDIMM-P solution. I also address the cost of fast nand in previous Blogs

Typically it is SLC with design modifications to speed reads and writes. Samsung and other NAND companies have indicated that SSD latency can be reducted to 10uS range with fast NAND

What specifics YMTC announces and how it will impact the NAND and SSD market will be discussed at Flash Memory Summit #FMS2018. Call us for more information

Link to PR announcement

Mark Webb

Wednesday, July 18, 2018

Intel and Micron to end 3D Xpoint Partnership

Intel and Micron to End 3D Xpoint Partnership

Micron and Intel are announcing that they will end their joint development efforts on 3D Xpoint . This is the latest update to the Intel/Micron Partnership and IM Flash

Intel and Micron will develop technologies independently after the second generation which is due out in 2019. All current products are first generation

Some thoughts 

Intel and Micron continue to separate business. NAND development partnership has ended and now Xpoint partnership will end quickly. The Fab is still jointly owned but Micron has options and has begun to document procedures and financing for ending the JV Fab agreement in their 10Q report.

side note: once you decide to not work together in the future, you stop working together in the present. Most likely every development and exchange of information will include a pre-analysis by attorneys. So it is safer to say that Intel and Micron are competitors and not partners today.

Intel is losing money on 3D Xpoint due to sales being below forecast, and costs are higher than the sales price. Micron is losing money on 3D Xpoint due to underload charges as they do not have a product and Intel is no longer buying from Micron as Intel doesn't need the output due to the previously mentioned poor sales

This decision is good for Micron as Micron has multiple options if Xpoint takes off or not and no anchor of future spending. Micron can buy out Intel from the Fab in Lehi at any time after Jan 2019 and run Xpoint, NAND, DRAM, or sell stuff to Intel. Micron is covered with options for either an upside or a downside in the 3D XPoint market and has leverage over Intel. Options are good

As publicly stated, Intels goal is to improve storage and memory capacity in datacenters using Optane  products and NAND SSDs. Intel can provide unique solutions that give Intel a competitive advantage on all servers. If Optane ends up in 50% of servers, the market is huge.... but we are not there (or even close) yet. If it becomes ubiquitous (a term I learned working at Intel), Micron wins by be a licensed source for Xpoint and with the partnerships Micron has in server world. 

Another result for Intel is that Intel can win as a company selling Optane at a loss (more processors, more servers, bigger moat). Micron cannot. But Intel needs to significantly improve sales to make this worthwhile

We will see what the possible next announcements are:

1) Lehi Fab decision ?

2) Micron announces a new NVM technology that they have been working on independent of Intel ??? (like when they announced their charge trap NAND)

3) Intel announces new partnership for developing and ramping Xpoint?

OR this could all get reversed.... a year ago Toshiba and WDC were in lawsuits, refusing to talk, locking engineers out of Fabs, and generally shaming eachother in public. Today they are one big consortium owned NAND Fab family. Maybe my friends at Micron and Intel will reconcile!

We will present technical status of XPoint revenue, endurance, speed, and configuration at FMS next month.

#optane, #Flashmem


BOISE, Idaho, and SANTA CLARA, Calif., July 16, 2018 – Micron and Intel today announced an update to their 3D XPoint™ joint development partnership, which has resulted in the development of an entirely new class of non-volatile memory with dramatically lower latency and exponentially greater endurance than NAND memory.
The companies have agreed to complete joint development for the second generation of 3D XPoint technology, which is expected to occur in the first half of 2019. Technology development beyond the second generation of 3D XPoint technology will be pursued independently by the two companies in order to optimize the technology for their respective product and business needs.
The two companies will continue to manufacture memory based on 3D XPoint technology at the Intel-Micron Flash Technologies (IMFT) facility in Lehi, Utah.
“Micron has a strong track record of innovation with 40 years of world-leading expertise in memory technology development, and we will continue driving the next generations of 3D XPoint technology,” said Scott DeBoer, executive vice president of Technology Development at Micron. “We are excited about the products that we are developing based on this advanced technology, which will allow our customers to take advantage of unique memory and storage capabilities. By developing 3D XPoint technology independently, Micron can better optimize the technology for our product roadmap while maximizing the benefits for our customers and shareholders.”
“Intel has developed a leadership position delivering a broad portfolio of Optane products across client and data center markets with strong support from our customers,” said Rob Crooke, senior vice president and general manager of Non-Volatile Memory Solutions Group at Intel Corporation. “Intel Optane’s direct connection to the world’s most advanced computing platforms is achieving breakthrough results in IT and consumer applications. We intend to build on this momentum and extend our leadership with Optane, which combined with our high-density 3D NAND technology offers the best solutions for today’s computing and storage needs.”

Costs for 256Gbit NAND Chips

Costs for 256Gbit NAND Chips

Samsung recently announced that it has 96L NAND chips in mass production. Interestingly, the chip is a 256Gb device which is at the sweet spot of NAND demand but is too small in density to be optimal for 96L. This will make it so that Samsung’s 256Gbit chip doesn’t quite take advantage of the increased efficiency that 96L brings.

So the questions are:

  • How much cheaper is this chip than the 64L Samsung chip to make?
  • How do costs compare among the leading competitors
  • The graph below gives some interesting data of costs for each company. More to come.



  • I have all the numeric details and the individual companies called out and similar reports for QLC, 96L, 128L etc and all the densities from 256-1Tbit
  •  For the first time ever, I am breaking out Intel and Micron costs! They use the same design at 64L 256Gbit but the Fabs are quite different today so the costs are different (Singapore vs Dalian). Toshiba and WDC cost are not broken out (same fabs)
  • All products are 64 or 72L 256GBit devices except the Samsung one called out as 96L
  • Die Size, Yield, process efficiency, Fab ramp, string stack all weigh in on cost
  • Costs are estimated for July 2018 and July 2019


  • 256Gbit is sweet spot for NAND density
  • Samsung 96L is just about equal in cost to 64L today. It will be cheaper in the future as it matures. 
  • 256Gbit is too small a die to take complete advantage of 96L density.
  • Samsung is not the leader in cost at 256Gbit in 2018
  • Intel and Micron costs are different for 256Gbit.

·         Tell me what you think. We will have more of the details at Flash Memory Summit next month.
#Flashmem, FMS 2018, Flash Memory Summit

Mark Webb, MKW Ventures Consulting, LLC.