This post is a follow on to my last post on OEM and in some respects to the post before that- on SSDs. This post in turn would make a nice lead in to another post on X3.
SanDisk’s current success in OEM can be attributed to strength in Mobile. While cards are great, the strongest Mobile growth has been in OEM embedded. To hear SanDisk tell it, this strength in embedded can be directly attributed to Adaptive Flash Management (AFM).
Basically AFM is a term SanDisk is now using to describe its flash system design expertise. As the slide below from this year’s February Investor Day (ID) illustrates, system design is SanDisk’s core strength. While SanDisk is world class in Memory Fabrication, Memory Design, System Assembly & Logistics, and Retail & OEM, it’s System Design is the special sauce.
Yoram Cedar, SanDisk’s Executive Vice President, OEM & Corporate Engineering, made the following comments on the above slide:
“And Sanjay talked about the vertical integration which is so powerful in this company. We essentially have world class in memory fabrication, memory design, system assembly, retail and OEM. We are world in class at that.
But we are also very good in system design. We are also very good in system design. We can take any one of our technologies and adapt it to different markets. And make sure that whatever is the best technology- that we’ll provide it. The best cost-effective technology that we will provide. We’ll drive it to that market and make sure that we satisfy the needs of that market.
And that entails in this and especially as it is related to X3 and also X2, it drives us more bits, more performance and ultimately more revenue per wafer- more GMs per wafer. And that is really our core capability. As Eli said and I think he said in this meeting, is that not all X3 are equal.
Well they are not. Maybe they are equal [pause]- I don’t think they are even on the wafer level, but definitely when you put it into the product with our system capabilities that X3 becomes a very very competitive product in the market and will satisfy all the customer’s needs.”
Through system design, or AFM, SanDisk is able to get more bits (X3 & X4), more performance and ultimately more revenue per wafer. Through flash system design SanDisk is able to transform raw NAND’s poor performance to meet target application specifications.
One of the most interesting presentations from Investor Day was Dan Inbar’s in-depth presentation on Adaptive Flash Management. Dan is SanDisk’s Senior VP & GM OEM Mobile and Imaging within SanDisk’s OEM group under Yoram and came to SanDisk from msystems.
The following is based on notes cribbed from Dan’s presentation.
Adaptive Flash Management (AFM
The slide below, also from Investor Day, is a great graphic illustration of what AFM (purportedly) can do.
Through the magic of AFM (light grey five-sided graph-barely visible here), the performance of the raw NAND Capabilities (center solid ochre five-pointed polygon) can be expanded to meet the requirements of the Application Target Specs (red-edged mostly light toned five-pointed polygon).
The five points (clockwise) are: Cost per Bit; Random R/W; Sequential R/W; Endurance; and Data Retention.
The capabilities of the raw NAND will vary depending whether it is SLC, MLC (X2), X3 or X4. The fewer bits per cell and the larger the geometry, the closer the raw NAND capabilities of the flash chip will be to the target specs on all points except Cost per Bit, where the opposite is true.
The smaller the geometry and the more bits per cell, the cheaper the chip and the lower the Cost per Bit. The illustration above is based on X3 on 32nm. Cost per bit is great (close to Target Specs) while the rest of the performance leaves a lot to be desired.
I had been wondering why there is the toneless gap between Cost per Bit Target Spec and Cost per Bit possible. It finally hit me. This is SanDisk (very dry) humor- the customer always wants it cheaper- no matter how cheap it is- hence the gap.
The slide above, also from Investor Day, shows the raw NAND capabilities of 32nm X3 as well as 32nm X2 and 43 nm X2. Cost per Bit improves- from 43 nm X2 to 32nm X2 to 32nm X3 whereas Random R/W, Endurance, and Data retention all degrade- as expected.
Sequential R/W performance for raw NAND 43nm X2 and 32nm X2 are the same, but drops significantly with 32nm X3.
The slide below illustrates how AFM can bridge the gap between raw the performance of the raw 32nm X3 NAND memory (light blue five-pointed polygon) and the requirements of the application target specs (the five-pointed red line polygon).
SanDisk has coined three terms to describe the rough outlines of how AFM works: Smart Caching, SLC Emulation, and High Performance Mode.
SanDisk’s AFM uses Smart Caching to improve the performance of Random R/W and improve endurance.
Smart Caching creates an area within the flash “which is like a sketch pad for the user. The user meaning the handset- the OS, the chip set. Being able to use that as an area that is constantly using [being used for] small quantities of data, fast moving data that you constantly need to use.”
To improve Data Retention SanDisk’s AFM uses SLC Emulation: “But if you consider the flash and you take a small portion for the code- which at the end of the day, out of the whole capacity is a very small part- if you take that capacity and allocate to it a small quantity and you consider [configure] it to work as similar to an SLC mode, then you’ve got the reliability that you need. There is no issue. There is no reason to add another component, which would be the SLC or even NOR- if you go even further back. You save the cost of that component. Therefore you have a much more competitive solution.”
To improve the performance of Sequential Write SanDisk’s AFM uses a High Performance Mode: “ Working with the system. Understanding the environment that we are in. Not being just a dumb memory device. Understanding the environment that we are in and having a dialog with the chip set or with the OS or the application. Having that dialog allows us to do a lot of things that before we could not.
For example, when you connect your hand set to your PC in most cases the USB device, you want to do cycles [sideloading?]. but also try charging the device at that time.
So what happens is that you also can actually change the mode that you use in the memory- utilize more power- because power is unlimited when I am connected to the PC really. And change the mode of working with the flash and therefore do significantly faster sideloading that you can’t do if you are working in the regular mode. But that requires understanding of what you are actually doing now in the memory and what is the handset environment.”
Basically, on one flash die SanDisk can manage different areas differently and can do it dynamically in real time, based on what is happening in the handset, in the OS and in the chip set.
Two important stories related to AFM are X3 and SSDs. I’m not going to go further into either here. Suffice it to say that thanks to AFM, SanDisk says it can bring X3’s poor random read/write, sequential read/write, endurance and data retention up to target specifications for embedded applications. As for SSDs, AFM appears to be at the core of SanDisk’s SSD strategy. To date the big success is pSSD.
A friend told me recently that he had heard that SanDisk is in talks with Intel (INTC). He didn’t know anything more. His guess and my guess is that these talks revolve around L&R.
Recently I’ve also started thinking a supply deal and an INTC investment in SanDisk are additional possibilities.
L&R discussions are obvious. For all intents and purposes the INTC/SNDK cross-license agreement has expired. The INTC/SNDK cross-license agreement was signed in October 1995 and extended for the life of the patents granted at that time.
INTC won’t need to license the expired SNDK patents, but SNDK has a second generation of most valuable IP, including AFM.
Last year Samsung was faced with pretty much the same situation. Samsung toyed with the idea that they didn’t need to license again because of patent expiration. On further reflection Samsung apparently decided all those other SNDK patents were worth a lot after all.
My guess is that Intel has come to the same conclusion.
A supply deal and/or an Intel investment is a bit more of a stretch, but makes a certain amount sense. Intel has said that it wants to be an SSD leader , but really doesn’t seem to care that much about where the NAND comes from .
“It may not be essential for us to have our own NAND factories to build (flash memory). We could probably specify the product that we want and buy it from third parties”
At the time (Feb 2009) most assumed that this would mean that Intel would pass along its NAND designs to a contract fab such as Taiwan Semiconductor. While this may be true, it could also mean that Intel intends to pick up supply from existing NAND players such as SanDisk.
An investment would be a bit more complicated, but makes sense if Intel feels SanDisk’s 3D R/W is the real deal.