by Gennadiy Detinich
11/27/2010 | 02:51 PM
In early October the microelectronics standardization body JEDEC held a series of seminars about the prospects of the flash memory market in general and of solid state drives in particular. Flash memory means much more to computer users than memory cards or USB drives. SSDs are not only trendy but also practical. The global economic recession has delayed the arrival of inexpensive flash-based storage devices but specialists do expect them very soon. How soon, exactly? Let's hear what the involved people, the developers of standards, technologies and end products, have to say about that.
In this article we will present the main facts extracted from the keynotes delivered at a JEDEC seminar to show you the big picture of the development of flash memory and its use in solid state drives.
One thing that we must keep in mind is that computers are not the most important application of nonvolatile memory. Flash-based storage devices in all their varieties are actually among the least popular products that use such memory, being only ahead of the car industry in this respect. Thus, the market in general cannot be affected much by how popular SSDs are or will be.
According to the chart built by Gartner’s analysts and the marketing folks from Phison, a leader in NAND flash production, the main driving force of the flash memory market is digital cameras (i.e. memory cards for them), MP3 players, USB drives, camcorders and mobile phones.
This situation is not expected to change in the next few years while the average annual growth in terms of memory amount will be about 100%. In other words, the amount of the world’s flash memory will be doubling each year from 2010 onwards. Sounds impressive.
The flash memory production graph only shows one slump, in 2008, over the last decade. Obviously, that’s the consequence of the global financial crisis. But regardless of the crisis, NAND flash has been developing much better than the other types (NOR, DRAM, SRAM), sporting a 40% annual growth throughout the decade (according to Forward Insights):
This rush has been due to manufacturing process improvements. You may know that flash memory chips are now produced on 24 to 27nm process whereas DDR3 SDRAM has only reached 30 nanometers.
When new technologies are introduced at such a tremendous rate, huge investments and high manufacturing costs are unavoidable. While we were pleased with the declining prices of SSDs, there was a negative side to the growth of the NAND flash market as the companies were suffering losses from manufacturing for tens quarters in a row! The crisis had nothing to do with that. The players just wanted to gain their ground on the new market and were ready to put up with temporary losses for the sake of long-term profits in the future. The NAND manufacturing industry became unprofitable in the second quarter of 2007 and only returned back to making profit in the third quarter of 2009. The total loss of the companies, excepting Samsung, was as high as $23.4 billion (according to Samsung’s data):
Samsung doesn’t report its own performance in the diagram and doesn’t reveal the identities of the other companies, but its results wouldn’t change the general trend much. The reduction of the price of flash memory chips (and SSDs) was the outcome of tough competition no company could avoid. In early 2007 the joint venture of Intel and Micron stepped onto the flash memory scene, signaling for others to make haste. The long-time business partners Toshiba and SanDisk began building new factories one by one. Capital investments were really huge.
It is in Q4 2008 that we saw the price of flash-based storage drop the most. That was the quarter when the industry almost came to a halt. For example, Micron stopped its NAND production in Idaho and the rest of the companies followed suit, provoking a shortage of chips and a growth of SSD prices. Fortunately, we’ve got a different situation now. There are new investments into manufacturing in 2010, leading to a new round of technological wars and giving us a promise of affordable SSDs (Gartner’s data):
Before the crisis, the average price of one gigabyte of flash memory had been declining at a rate of over 50% annually (by 64% in 2008, for example). Starting from 2009, this rate is lower at below 40% (Forward Insights’ data):
This year the average price of one gigabyte of NAND flash is a mere 18% lower than in 2009 which has brought us to the same situation as in Q4 2008. However, the price reduction is going to be 38% in the next year, reaching the important level of 1 dollar per gigabyte. Running a little ahead, the dollar-per-gigabyte price is expected in 2012 for SSDs.
The key factors determining the flash memory pricing are the increasing production volumes, the transition to 64-gigabit chips, and the growing share of NAND MLC memory with 3-bit cell. However, the rather low demand for SSDs doesn't help to speed the process up. The low popularity of SSDs is due to both users who don't want to pay much more compared to traditional storage devices and to the manufacturers who should make more promotional effort to tell about the benefits of SSDs and come up with a single standard for evaluating their consumer properties.
For example, the mean-time-between-failures parameter is absolutely uninformative for SSDs because NAND flash has a limited number of erase cycles, and each manufacturer has a different approach to specifying an SSD’s number of allowable rewrite cycles and speed. Fortunately, the JEDEC JC64.8 SSD Endurance and Data Retention Verification Test we will discuss below is going to make it easier to compare the specs of flash-based drives from different makers.
Talking about the distribution of nonvolatile memory among different device categories, Samsung envisions dramatic changes in the driving forces of the market. In this year memory cards and USB flash drives begin to lose their leading position.
Smartphones, SSDs and tablet PCs are the new demand-boosting factors. With Apple's endorsement, tablet PCs have given rise to a whole new segment in the computer industry and promise to grow threefold, from 3% to 9%, among the key NAND flash consumers. SSDs will also grow impressively from 6 to 10%. Smartphones will increase their share by 4%, which is quite a lot, too.
Overall, a total of 9.5 billion gigabytes of NAND flash will be produced in 2010, which is 74% more than in 2009. That’s not the biggest annual growth in the last decade, though. This memory type made its biggest leap in 2005 – 245% in one year! – to leave NOR flash behind in terms of production volumes.
The expectedly high demand for tablet PCs and the growth of the smartphones’ share are the consequence of the rising popularity of social networks and wireless communication devices. For example, the Facebook social network grew from 250 to over 500 million active users over only one year from 2009 to 2010. The daily number of its users has increased from 120 to 200 million. The number of photographs posted on the site’s pages has increased, too, from 1 to 3 billion. In 2009 there were 10 million monthly updated videos, but now this number amounts to over 12 million. Interestingly, 120 million users have mobile access to the site whereas one year ago mobile Facebook users were only one fifth of that number.
The numbers are indicative of a significant growth of transferred multimedia data and of the ongoing shift towards mobile devices. Notebooks are not necessary and even redundant for consuming the informational noise generated by social networks whereas smartphones and tablet PCs are quite enough. Both these categories of mobile devices utilize flash memory. You won’t find HDDs in smartphones, you know. However, it is necessary to quickly increase the amounts of memory in such devices to keep up with the customers’ growing demands.
The important role of social networks in the regular exchange of Internet traffic is indicated in a study conducted by Morgan Stanley Research. As the diagram shows, the number of users of various Web communities exceeded the number of registered email users in July 2009. And it is as far back as November 2007 that the total time Internet users spend communicating in social networks exceeded the time they spend on email.
By the way, these diagrams provide some ground for analysts to predict a death of email communication in the next few years. By the end of the next half a decade email will remain only as a work tool in business applications and will account for but a very small share of an average user’s time and traffic.
According to Cisco Visual Networking Index, the average monthly traffic is 21,367 petabytes in 2010 (1 PB equals 13.3 years of Full-HD video). With a predicted average annual growth rate of 40%, the monthly amount of traffic will have exceeded 600,000 petabytes (or 600 exabytes) by 2020. By 2040, the monthly traffic will be as huge as 500 zettabytes. We will only have yottabytes left for measuring as we haven’t invented larger measurement unites as yet. We need a new unit, adds Samsung jokingly.
Driven by Apple’s will, tablet PCs may sell 15 million units in their first year on the market. This will largely be Apple’s iPad since most competitors won’t be in time to release their alternatives even by the Christmas season, yet we are still looking forward to such products as the Samsung Galaxy Tab, ASUS Eee Pad, Dell Looking Glass, HP Slate, Motorola Everest and Cisco Converj. The total sales of tablet PCs are going to increase to 50 million units in 2011 and to 80 million units in 2012. The diagram from iSuppli shows the sales growth and the amount of flash memory used, indicating that the amount of that memory is growing up at a faster rate than the number of tablet PCs. The diagram from Forrester Research shows the share of tablet PCs on the USA personal computer market.
We guess these data are going to be roughly the same for other developed markets, i.e. for Western Europe and Japan. You can see that there is now a fourth element in the PC market starting from 2010. And this element is now only growing by itself but also eating up the shares of the older PC categories. Analysts from Goldman Sachs Research predict that in the years of 2010 and 2011 about 60% of all tablet PCs sold will be demanded for their own sake whereas 40% will be bought to replace notebooks and netbooks. More specifically, 95% of those replacement tablet PCs will replace netbooks and 5% of them will replace notebooks. The PC market will shrink by 7 million systems as the result.
In 2011, the percentages will change: 50% of the buyers of tablet PCs will prefer them to netbooks and 20% of the buyers will prefer tablet PCs to notebooks. The resulting 20 million systems will add to the market of flash memory which remains the only reasonable technology for tablet PCs and is not used much in netbooks or notebooks. An increase in tablet PC sales by 3.7 times will provoke a 5-fold increase in flash memory consumption because of the ever-growing amounts of digital content.
Another notable contributing factor to the growth of the flash memory market is the market of electronic readers. Although some people argue that the Apple iPad is going to kill the e-reader as an independent class of products, this market is expected to develop at its own rate and consume its share of NAND flash, increasing the amount of onboard memory per device.
The average annual growth of e-reader sales is expected to be 70% in the next few years. The online shopping site Amazon already sells more electronic texts than hardcover books, expecting to outperform the conventional paper in every type of publication during the year. The Amazon Kindle is popular because the e-reader itself is priced close to its manufacturing cost whereas the manufacturer collects its profit by selling texts.
Finally, there are smartphones which will also show a steady sales growth of about 36% annually in the next two or three years. In fact, the mobile phone market at large has been expanding only through smartphones in the last two years while the sales of ordinary mobile phones stay at the same level. The average amount of integrated NAND flash memory in this year’s smartphones is 5.9 gigabytes. This number is going to increase to 9.5 gigabytes in 2011 and to 13.3 gigabytes in 2012, etc (data from Forward Insight). Considering that users often expand their smartphone memory with memory cards, this category of devices is increasing its consumption of flash memory at a high rate, too.
Today, SSDs are rather too expensive. And if you can find an affordable SSD, its characteristics are likely to disappoint: a $100 flash-based drive is going to have a small capacity of 32-64 gigabytes and a rather low performance. The manufactures try to convince us that the cost of storing one gigabyte of data is not the most important factor for SSDs and that we should appreciate such factors as reliability, shock resistance, high IOps per dollar, low temperature, lack of noise, etc. A corporate user will be willing to pay for all this but a home one will be waiting for a lower price.
Both types of users need an individual approach, but the general trend is to replace SLC NAND flash (with 1-bit cells) with MLC NAND flash (with 2-bit and even larger cells). The enterprise MLC type also comes to the market, boasting an increased number of rewrite cycles (6000 instead of 3000). Intel uses such eMLC memory in its new enterprise SSDs. This is all meant to reduce the price of the end product, of course. But SLC-based drives will still be used where the price factor is unimportant, like data warehouses with a huge amount of requests.
In 2009, the cost of storing one gigabyte of data on an enterprise-level SLC-based SSD was up to $20. In 2010, thanks to the release of enterprise-level SSDs based on MLC memory, this cost is as low as $5. These SSDs are intended for servers. SLC-based SSDs are still preferable for large data storage arrays. However, in the next year MLC (or maybe eMLC) based SSDs are expected to span the entire corporate market and replace SLC-based counterparts in many of its sectors, lowering the cost of storage to $5 per 1 gigabyte. As a result, we can expect the price of 1 gigabyte of a consumer SSD with MLC memory to drop to $1 in 2012 (corporate SSDs are going to notch the $1-per-gigabyte mark in 2014).
The following 5-year forecast from Gartner (made in August) shows the development rate of the different sectors of the SSD market.
The majority of solid state drives – about 90 % of the total – are going to target client systems including home desktop and mobile PCs and netbooks. SSDs will be obligatory in top-performance home computers and workstations. They are also going to become an indispensable component of slim notebooks. Less advanced computer systems will either use SSDs optionally or together with HDDs in a hybrid storage system where the OS and applications are loaded from the faster SSD whereas data are stored on the slower HDD. Netbooks will have SSDs by default, but in small quantities. This is Samsung’s vision of the SSD distribution map depending on the cost of the computer system:
Getting back to the Gartner diagram, we can note that the company had lowered its forecast for the SSD market growth. Compared to the July version of the document, the difference is 5 to 10 million units annually. We can suspect that the analysts had corrected their forecast due to the rapid development of tablet PCs which, as we’ve seen above, eat up a share of the PC market. Anyway, the average annual growth of the SSD market will be higher than 70%. This is more than what is expected for tablet PCs, so computer applications of flash memory are still going to be an important factor for the development of NAND flash.
It is especially pleasant for us, home users, that even though consumer SSDs will account for 90% of the total SSD sales, they will only bring their manufactures half the profit. The other half will be provided by the buyers of enterprise SSDs (Gartner).
Now here are two more graphs to show you when the total amount of flash memory is going to exceed the total amount of HDD storage and to evaluate the price reduction rate for SSDs.
The diagram on the left is from IDC. It promises that SSDs will exceed HDDs in terms of the total amount of storage in 2018. The graph on the right, based on Samsung’s data, shows how retail prices on consumer SSDs are going to decline. 64GB SSDs have already crossed the price mark of an average HDD plus $100. 256GB SSDs will cross the +$100 mark in three years. 1TB SSDs will reach that mark in 2016.
While getting cheaper and cheaper, SSDs will also be improving in terms of their consumer properties, which are already quite enticing. The manufacturing technology, interfaces, controllers and the very operation principle of SSDs are all going to be polished off.
Right now, the replacement of SDR with DDR signaling should double the bandwidth as it was with the transition from SDR to DDR SDRAM. As a result, the speed of data transfer will increase from 66 Mbps to 133 Mbps and higher, up to 200 Mbps.
Like with system memory, DDR flash is going to be synchronized with an external DQS signal. There are two possible operation modes: asynchronous (toggle mode) and synchronous. Without delving deep into the technical details, we can note that the first mode is used by Samsung in its newest flash memory chips. The synchronous mode described by the ONFi standard (Open NAND Flash Interface) is promoted by Intel with its partners. DDR flash memory with frequencies up to 200 Mbps is going to be ratified by JEDEC in early 2011. The next step will be to increase the speed of the flash memory interface to 400 Mbps.
Samsung illustrates the benefits for SSDs from transitioning to DDR chips with 133 Mbps and 400 Mbps frequencies.
The number of IOps in random read mode is going to grow by 10% and then by 20% whereas the number of IOps in random write mode by 40 and 60%, respectively. The current generation of NAND flash can only provide the first step of the acceleration. Transitioning to the high-speed 400Mbps interface will call for a larger signal structure and for the introduction of voltage arbitration signals, providing the option of external power supply for memory cells during read and write operations.
The efficiency of the integrated power source of NAND flash, which converts 3.3 volts to 11.5-12.5 volts, is less than 30%, which is especially important for server-oriented SSDs. On-die termination will also have to be introduced. It will be manageable, just like the strobe latency (you can refer to this document for details). The new signals will only add to the existing ones, making the high-speed memory chips backward-compatible with old controllers.
As another important improvement in NAND flash, the supply voltage is going to be reduced from 3.3 to 1.8 volts. This will coincide with the introduction of the first-generation DDR flash (133 Mbps). And the next step will be to increase the number of allowable erase/write cycles in enterprise MLC memory. Today’s MLC flash memory type supports 3000 rewrite cycles whereas early commercial versions of eMLC increase this value to 6000 (in the new enterprise SSDs from Intel). The developers even promise to improve eMLC to 30,000 rewrite cycles in the future.
To understand the ways for SSDs to improve, we can take a look at the data from Qualcomm showing the changes in the basic specs of flash memory.
Currently, the leading manufacturers including Intel, Micron, Samsung, Elpida, Hynix and Toshiba are preparing to switch from 30nm to 20nm tech process for NAND flash. Alas, the smaller tech process will worsen the consumer properties of flash memory cells. The service life is reduced from 3000 to 2000 or an even lower number of rewrites (we are talking about the MLC memory type here as the other types won’t switch to the 20nm tech process). The charge retention time is going to decrease, too, but not far below 5 years, so that’s unimportant for SSDs which are used on a regular basis.
The size of a data page may remain the same at 8 kilobytes but the manufacturers will hardly be able to keep the raw bit error rate on the same level. As the tech process gets thinner, the interference of adjacent cells grows up, the number of electrons in the floating gate is decreased, and the leakage currents in the slim insulators get higher. All of this calls for more ECC bits and for faster ECC processors. The existing 24 ECC bits for 1 kilobyte of data won’t be sufficient for long. The 20nm generation of chips will have 80 or even more ECC bits per 1 KB of data. Besides, the controller developers suggest adaptive ECC which would be able to occupy memory pages intended for data.
The problems provoked by the increased error rate, lower service life, larger memory page size, and larger MLC cell (3 or even 4 bits) all require more effective ECC units and make the design and operation of the flash memory controller far more sophisticated. Maybe this is why the backward compatibility of future high-speed NAND chips with existing controllers is so important?
The controllers will have to get smarter as the result. For example, some section of a memory chip may be defective. It may be a line or a column or something else. Factory testing doesn’t always reveal such problems whereas a deeper test would be more expensive. SandForce, a manufacturer of flash memory controllers, suggests that future controllers are going to be equipped with a system of fault tolerance. RAID arrays are the best mechanism of fault tolerance available today, so future SSD controllers will have to incorporate on-die elements of RAID architecture.
As another notable improvement in controllers, the internal SRAM and external DRAM buffer are going to be replaced with internal nonvolatile memory (SandForce has already given up an external DRAM buffer in its controllers). The nonvolatile buffer may be based on such perspective flash memory types as resistive ReRAM (the so-called memristor), magnetoresistive memory (MRAM) or phase-change RAM. As opposed to NAND flash, these memory types are considered “eternal”. They are also fast enough to serve as buffer memory. Of course, such a buffer wouldn’t be affected by power failures. In the long-term perspective, not only the buffer but also the whole SSD can be transferred to new memory types such as the mentioned three.
There is also a concept of burning NAND flash cells with variable speed to achieve more flexibility in the operation of SSDs. The concept of adjustable dynamic and static performance allows producing SSDs tailored specifically to different applications. If you need the maximum speed even though the service life is short, you can use static performance. If you need the maximum service life and want the peak speed in some situations only, you can use adjustable dynamic performance.
Thus, the flash memory cells will slow down in noncritical tasks but will offer their best performance when you need it.
Of course, a practical implementation of all these innovations requires some improvements on the software part of the controllers, OSes and host controllers. Today’s computer platforms are HDD-oriented and, as our tests suggest, the developers of RAID controllers and chipsets still have a lot of work to do until SSDs begin to behave predictably. Here is an illustration from Samsung:
The same SSD behaves completely different on different controllers, the variation getting bigger as the request queue depth gets longer.
We’ve mentioned new types of nonvolatile memory above. It must be said that NAND flash and its ecosystem are going to constantly improve, too, in terms of interfaces and tech processes as well as cell depth, packaging and microarchitecture.
Flash memory with 3-bit cells has occupied up to 25% of the NAND flash market in the last 2 years. It was first produced by Toshiba but now the Japanese company has been joined by Samsung and Intel with Micron. By the end of the next year 3-bit-cell memory is going to account for 35% of the market whereas 4-bit-cell memory will take 5% of it.
3- and 4-bit-cell flash is supposed to have different applications: memory cards and USB drives, respectively. The higher bit depth of a memory cell makes the controller and ECC much more complex. Although such memory can lower the price of storage, it has lower performance and a lower number of rewrite cycles, but we have no doubt the manufacturers will try to find a compromise between low price and acceptable reliability and will eventually produce SSDs with 3- and 4-bit-cell memory.
Another problem is the growing number of contacts of NAND flash chips which can be as high as 600 pins for the highest-capacity chips. This is a problem for the manufacturers that want to make the chips smaller. The solution is in improving the system-in-package and package-on-package technologies as well as in introducing the through silicon via technology when holes are created in the die with a laser and then filled with copper.
Transitioning to serial interfaces may also be a good solution of the problem of the large number of pins. As a replacement for embedded flash, the first generation of flash with a serial interface is going to come out in the next year. It is called Universal Flash Storage:
UFS memory will not only make the data-transfer rate higher but also change the operation logic of flash drives in Internet gadgets, smartphones, etc. The UFS standard supports multi-tasking, creates a flexible request queue (borrowed from the SCSI concept), scales up from 300 MBps and higher, and allows building a chaining topology to reduce the minimum required number of the controller’s or even the host-controller’s channels. In fact, the UFS standard may grow up to be a full-featured computer interface rather than an interface for embedded NAND flash only. In fact, NAND UFS is something like single-die SSDs.
Despite the ratified JEDEC specifications (JC-64.8) there are still no commonly accepted methods of measuring the reliability and other consumer properties of SSDs. It is impossible to compare products from different brands, which confuses the consumer.
So far, the following things are certain. All SSDs are divided into two classes: client and enterprise. Each class has its own requirements. For example, the characteristics of client SSDs will be correct for an operating temperature of 40°C and for a daily operation of 8 hours or less. Enterprise SSDs must retain their specified properties during all-day operation at up to 55°C. The allowable data retention (in the turned-off state) is 1 year for client SSDs (at 30°C) and 3 months for enterprise SSDs (at 40°C).
This means that the controller has to rewrite, from time to time, even cells that are not written to. For example, it has to renew data in read-only archives and do so four times more frequently in enterprise systems than in client computers. The functional failure rate for both types of SSDs is not higher than 3% while the UBER parameter is 1015 and 1016 for client and enterprise SSDs, respectively.
There will be a different system of ratings for each product class but it will be based on the TBW (terabytes written) value, a decimal value close to the 100% wear of the specific SSD under the specified conditions and at the allowable level of errors.
The service life of an SSD is measured directly by writing until the SSD fails due to the exceeded limit of rewrite cycles or is calculated by extrapolation. In the latter case we have to rely on the manufacturer’s honesty in performing the math.
For today, the JEDEC standard only defines the operation load for measuring the service life of enterprise-class SSDs. It is based on synthetic benchmarks rather than on emulating some real-life applications. The whole user-accessible capacity of the SSD is loaded at that. The load for client-class SSDs is still under development. All the specifications are provided for 512-byte sectors. The load and rating system for 4KB sectors is under development.
Summing up the facts presented in this article, we can give you the following synopsis. The development of social networks and related services provokes users’ interest in multimedia-supporting digital devices which need rather large displays and amounts of memory. Tablet PCs suit this purpose very well, so they have a chance to capture a share of the notebook market as well as to create their own market niche, stimulating the growth of the flash memory and SSD markets. In two years the cost of storing data on SSDs is expected to drop to $1 per gigabyte. Moreover, the 2012-model-year SSDs will be much faster than today’s $100 products. And in six years more, SSDs will exceed traditional HDDs in terms of total storage space. It is going to be very interesting to watch the new technologies and devices emerge on the market during that process. Stay tuned to us, as there’s a lot of exciting things ahead that need to be tested in X-bit's labs!