Plotting the future of memory
Electronic News and Electronic Business sat down to discuss changes in the memory market with Mueez ud Deen, director of marketing for graphics, mobile and consumer DRAM at Samsung; Ali Mesri, director of OCT product management at Qualcomm; Russell Barck, director of business development and strategic alliances at Spansion. What follows are excerpts of that conversation.
Q: What will win in the memory space? Will it be DRAM, NOR, NAND, or some other variation?
Deen: There isn’t a battle. These products are complementary in nature. DRAM has been and continues to be the glue between the processor and the hard drive. It can synchronize the speed of the processor with the hard disk. Flash will complement that. It’s one of the additional mediums of long-term storage.
Mesri: Our OEMs are users of both. The mobile space is a special category because there is a low-power version. This is not generic DRAM. And, of course, performance and flash also plays a significant role both for code and for data storage. We follow the trends and the standards, which creates an equal playing field for all the suppliers.
Barck: There are certain categories of the market where each has its strength. From an end user perspective, there isn’t any competition. From a product implementation, though, there is. Data permanence and reliability favors NOR, which can move into a critical role in the data storage marketplace. The higher-density products move us into the traditional NAND space.
Q: This is a crowded market. Who survives and who doesn’t?
Mesri: There may be too many suppliers now, but there will be a consolidation.
Barck: There has been a large consolidation in the market over the past several years. It is extremely competitive. The ones that will survive will be the ones that can drive leading edge technology into the marketplace and drive the manufacturing infrastructure to support that with low-cost products. It’s innovation coupled with large-scale manufacturing.
Q: Is this something that will be driven by low-cost labor or better use of technology—synergistic type of development. And also, what makes one type of memory more attractive than another? Is it cost?
Mesri: Back in the ’60s and ’70s, labor cost was one reason manufacturing moved from North America to Japan, and then to Korea and Taiwan. But the process of building these chips is automated. The factors that contribute to lowering costs are faster processes and the ability to minimize die size and maximize yield. Those are the cost factors that make a company shine, not labor.
Barck: Labor has become less of a factor. Having the manufacturing expertise and the fabs in place drive down the cost. The architectural element is also critical. Not all architectures are created equal. In the NOR space, the MirrorBit architecture is a more efficient architecture.
Q: Does the average person know that when they go out and buy a consumer electronics device?
Barck: I don’t think the average person knows that, but it translates into better performance and better price, and that’s something they understand. They don’t understand lithography, but they do get higher reliability, higher performance and price.
Q: But even for the technologically savvy buyer, there are so many tradeoffs in a design and factors that can influence performance that it’s hard to tell what’s worth paying for and what isn’t.
Barck: It comes down to an optimal consumer experience, and there are all sorts of ratings in the industry in regards to performance.
Q: Let’s look at this from the OEM standpoint. Does Qualcomm see a difference between one product and another?
Mesri: There are definitely differences from supplier to supplier. There are also differences by platform. Different categories of platforms, from value to convergence, use different configurations of memory for the code and storage. For the lower-end platform, we use lower-speed, lower-performance. It’s also possible to use a similar configuration at the higher end, but we use much higher performance and throughput with DDR or SDRAM configurations. I do agree the customer doesn’t’ understand the advantages. From their perspective, all they know is that the price goes down. If a gigabyte costs $10 this year, it will be $5 next year. But OEMs do know the products. They know the specifics from one market to the next.
Q: Samsung turns out a lot of different cell phone models every year. Does each one have a different memory configuration?
Deen: Not necessarily, but there are at least four to six different categories of memory. A low-end GSM product built for areas that are growing and only need voice is very different from a high-end, full-multimedia phone sold in Japan.
Q: The memory market has been prone to huge inventory swings, as evidenced by numerous charges of dumping and price fixing. Is that changing?
Deen: Obviously no vendor wants inventory. Usually excess inventory occurs because you overshot the market. Right now, some memory is in ample supply while there are shortages of other types. The key thing that differentiates it is the end market. Flash is driven typically by non-PC type applications. DRAM is very PC-driven and cell phone-driven. If there is an inventory problem, something happened that was beyond their capacity to control.
Barck: The industry has definitely gotten better at managing inventory. As you go toward worldwide demand, you establish numerous regional warehouses that can satisfy customers. The trend is moving more and more toward build to order, where our customers are moving to reduce their inventory, as well. We get demand forecasts from customers, and build it for just-in-time delivery so there aren’t large warehouses sitting full.
Deen: The fabrication process takes 12 to 14 weeks between the time the process goes in and the wafers come out. When you make a commitment and something changes, what do you do? You get the wafers in three months and you have to deal with the inventory problems. It’s hard to react in a short timeframe. That depends on fab efficiency and diversification of the product line. Depending on where the market is moving, you can slow down some of the wafers and divert some capacity.
Q: Is this a matter of you get burned and then you learn from that mistake? Inventory problems have cropped up successively over the years in the United States, Japan, Korea, Taiwan and now China.
Deen: It’s a case of live and learn.
Mesri: At Qualcomm we don’t deal with the inventory. The OEMs do. Flash is in short supply in some areas right now. The memory usage and the memory footprint on some platforms like MP3s and cell phones are going up and up. There is capacity, but keeping up with it is difficult.
Q: What will drive memory sales in the future that isn’t there today?
Barck: The technology will change, but the fundamentals behind those applications won’t change. In a cell phone, for example, how fast does your phone boot? When you get into gaming, performance is very important. So is the richness of the experience, and that’s driven from a data perspective. It’s also a performance element of the battery life and the ability to extend that over multiple days. One of the other areas that will become important with applications like mobile TV and downloads is security. The ability to tightly couple security with the content will become more important.
Deen: The future for memory is expanding, and it is virtually unlimited. It allows a richer users experience, a more convenient user experience, and it comes at an affordable cost. Cost is also very important in making it approachable for the consumer.
Q: Isn’t cost the underlying factor of all of these?
Deen: Not always. Sometimes it is performance. In graphics cards, we introduced a fourth-generation memory chip. Its sole purpose is to provide a high-performance interface that will provide a realistic environment. The cost is a secondary factor. In the PC side or even mobile DRAM, the main factor is how little power does it take?
Mesri: With higher data throughput and 3G technologies, higher uplink and downlink speeds and throughput and connectivity, and 4 gigabyte cards for high-end cell phones and high-definition movies that can take upward of 4 gigabytes, these applications drive the improvements on the phone. There are products from the silicon vendors with 4 gigabytes on a single die. That was not achievable three years ago. Two years ago the requirements for a cell phone were well below a gigabyte. Now phones are being designed with 1 gigabyte of flash. Vista will require 4 gigabits on a cell phone so you can run the majority of your applications on your phone.
Q: Memory used to be an entry point for customers, but the investment necessary to get into the market now is much different. Will that limit the number of new players coming into the market?
Mesri: Technology also matters quite a bit. Performance is important. So is power. You can put a high-performance memory into your cell phone, but it drains your battery in 15 minutes. As you migrate to more advanced processes, it becomes more difficult to use the same processes and come up with the same features. But the OEMs want these same features, so it becomes very challenging for the manufacturers. Performance, endurance and reliability need to remain the same as you migrate down to the next node.
Barck: This is one of the reasons for the consolidation in the market. It’s very expensive and it’s very capital-intensive. Semiconductors have been capital-intensive for some time. Everyone is going fabless right now because of that barrier to entry. Our closest competitors are losing hundreds of millions of dollars a year. The ones that will survive will have a balance of technology coupled with performance and market share. As we go forward, lithography node transitions are happening faster and faster. You need scalability in your architecture at advanced lithography nodes.
Deen: Most startup companies seem to be able to get access to funds no matter what. In certain countries, the government becomes a benefactor and supporter. Easy access to money has prevented a major consolidation even though the cost of equipment is rising. If you look at anyone deciding on building a new fab, they have to take billions of dollars out of profits. That’s a big challenge. When people go fabless, that doesn’t consolidate the industry. It just means you’ve transferred the responsibility. The bottom line is whether consolidation will mean a reduction of supply. I don’t believe that has occurred and I don’t see it happening in the near future. But who can actually handle production at 45 nanometers? That requires experience, expertise and engineering resources.
Q: Has anyone started producing at 45 nanometers?
Deen: There is no DRAM, but we do have flash at 45 nanometers. Our goal is to get to the next node early so we can get a head start on solving the problems. That’s the way we like to work it.
Mesri: We definitely have product in production at 65 nanometers. We have good visibility into the advanced trends. All the suppliers are looking at advanced nodes, but there will certainly be challenges. I don’t think some of those challenges are clear at this time. There are challenges for us 45 nanometers, too, but this is something we have to do. We have no choice.
Barck: We’re at 65 nanometers and we’re migrating our latest 300mm fab to 45 nanometers next year.
Q: How far down the Moore’s Law road map go from your vantage point?
Mesri: We definitely have clear plans for the next migration. There’s a lot of planning in advance you have to do. We have a clear understanding of what will be required from the standpoint of design rules and the next node, 32 nanometers, but beyond that it’s not clear.
Barck: We all follow this year to year, and it continues to blow away expectations as to what is possible. Moore’s Law continues. Our corporate goal is one technology node a year. We have teams working on what’s below 32 nanometers. We are also working on new architectures that can get us beyond that.
Q: Such as?
Barck: We haven’t announced these yet. However, there are a number of startups that have announced interesting technologies in this area. Some of them are nanotechnology-based, some of them are new physics implementations of what we’re doing today.
Deen: We’ve been continually surprised. When we thought there was a wall, it moved. Wherever we perceive the wall to be, we are working on moving it further out.