August was a pretty slow month, however there were quite a few interesting things happening in the memory market. The extreme growth of the manufacturers’ production capacities, the leveling-out of the DDR1 and DDR2 prices, the developments on the DDR3 and the extreme growth of the Nand Flash application.
The leveling-out of the DDR1 and DDR2 prices, the aggressive push by Intel to adopt FB-DIMM as the next memory for server application and the latest developments on the DDR3 side and more in our quarterly newsletter.
At the recent JEDEX Conference held at San Jose on April 2005, T S Jung of Samsung declared the "Digital Storm" has started. Memory consumption will be greatly increased due to convergence of the computing machines into the consumer market. More information can be found at http://www.jedex.org/
At the IEEE Industry luncheon in June, Dr. Wang, CEO of Samsung Semiconductors, once again had emphasized the "Perfect Digital Storm". He forecasted memory usage on PC will lose its position to the other electronic appliances. PC memory consumption, will for the first time, will occupy less than 50% of the total memory production.
Nand ! Nand ! Nand ! everyone trying to get a piece of this market
On September 2005, Dr. Wang of Samsung, declared the "NAND Flash Rush" has started. Apple Computer announced that the new iPod Nano music machine can hosts up to 1 gigabyte of NAND memory. Followed with the rumor that the new Instant Boot Microsoft machine will place its operation system onto Flash memory. These two potential applications had created a fear of NAND shortage in the near future. Industry expert predicts NAND Flash production capacity will lag behind demand for the near term.
On September 2005 Motorola has announced the availability of a music cell phone allowing the download of 100 copyright title songs into the cell phone. The cell phone now overlaps the iPod and the MP3 music machine in the market. The latest camera phone has also incorporated the 3.2 Mega-pixel picture sensor. Next generation will definitely be increased to the 5 Mega-pixel and 8 Mega-pixel. Demand for picture storage memory (NAND) will increase with it.
Still pictures on cell phone will not satisfy the appetite of the tech savvy users. They will soon want motion picture and video streaming. Multimedia feature into the cell phone will be a requirement. Large blocks of DRAM memory will have to be used to fuel this feature. The new Low Power DRAM (LPDRAM) is created for this application. It is designed for low cost and low power consumption.
"MCP" ! a new goldmine for dram manufacturers
In order to better package these large memories into the hand phone, MCP (multi-chip package) memory is required. MCP combines various varieties of memories like Pseudo RAM, NOR, NAND and LPDRAM all into one package to save space. JEDEC, the memory standardization committee has been working on a unified standard for multi-vendor interchange ability. When the standard takes hold in a few years, MCP prices will drop and even larger blocks of memory will be used.
Recent announcement of the Blackberry handheld adopting Intel processor and the CE operating system is further evidence of technology consolidation in the hand held industry. Hand phone and mobile will soon be built on only a handful of standard platforms like the PC had gone through. This is purely for better affordability and mass-market adoption.
DDR versus DDR2
DDR2 memory will finally gain acceptance as mainstream memory in 2006. Intel and memory vendors attempted to switch from DDR memory to DDR2 memory back at the beginning of 2005. The effort failed due to their miscalculation of consumer sentiment. Consumer was not willing to pay more for small increase in performance. While the first generation of DDR2 400/533MHz memory were barely in par performance compared to the DDR400 memory, consumers were not willing to pay the $200+ for the new mother board that give them DDR2 memory, PCI Express video and SATA hard-drive. That was too great a price jump for the consumer to accept.
Instead, the next wave of low cost motherboard might not include the costly PCI express nor the SATA option just to lure the consumers into the new DDR2 memory. To get more performance from the DDR2 memory, memory data width will have to be increased to 667MHz and 800MHz. Unfortunately, 800MHz-memory performance can only be produced with 90nano-meter semiconductor geometry. 90nano-meter memory production would not be cost effective until mid-2006 when most of the DRAM vendors have completed their small geometry conversion on their production line. With AMD joining the camp to use DDR2 by summer 2006, DDR2 would have a "take-off" around mid-2006.
What is VLP DIMM ?
With digital video-on-demand and fiber-to-the-home becoming a reality, more servers will have to be provided to supply the continuous video streaming source. Mid-end server market will grow. More memory for these servers will be required. Intel is proposing the new Fully Buffer DIMM memory system to fill this requirement. On the other hand, the memory industry is proposing lower cost alternative solution like registered DIMM, 4 ranks DIMM and VLP. These VLP memory are memory modules designed in MiniDIMM VLP (Very Low Profile) format, which were only 18.3mm high.In this battle, low cost and reasonable performance will ultimately gain the acceptance of the consumer.
On the memory production side, some premier memory lines will be converted to NAND production to fuel the "RUSH". That would keep price stable for the first half of 2006. DRAM production would gradually shifted to the Far East. Subcontract manufacturers in the Far East like SMIC, Grace Semiconductor, Winbond, Nanya and Powerchip will see exceptional growth in their dram production.
Fabless Trend , a new era for manufacturers
Untested NAND and DRAM wafers will become a norm, as distributors are desperate to find innovative ways to cut cost. More and more DRAM and NAND memories will likely be test in the final product instead of at the chip packaging level. Because of that, some memory vendors may abandon their production business. Instead, they will become fabless. Vendor consolidations will likely take place.
The basic DRAM cell has not been changed since the first 1T DRAM cell was invented. For the last two decades, the changes were in the system architecture and more innovative ways to squeeze the last bit of performance. All these will come to a diminishing return. Higher speed memories will not be able to compensate for the lost in latency. New concept memory will have to be found. Many Nano-memory and atom level memory cells have been demonstrated by scientists. Engineers will eventually have find a way to make them cost effective to produce. We will then see memory growth on exponential curve again.
In summary, it is quite difficult to make any definite long term forecasts of memory trend these days. On one hand, DRAM manufacturers keep expanding their production capacities in anticipation for the holiday seasons and yet realistic demand have not peak yet. This have resulted to weak demands for PCs causing the more expensive DDR2 prices to dip below DDR1 memry prices.The recent "Hurricane Katrina" devastation and the high energy cost is causing consumer to spend less at the retail store , and its unclear if this Christmas season will be jolly.