Monday, May 06, 2002
It has been over a year since I wrote "Memory Trends 2001". The landscape of the memory industry has changed a lot. I am, therefore, enticed to write this new article to cover the interesting things I have seen for the past year. Hopefully, we can see the future through the crystal ball with help from history.
The Great DRAM Price Collapse
While price of PC memory has been going down, the megabyte has been going up. As recent as 1995, mainstream PC memory was selling at U$1 per megabyte. Due to the shrink in semiconductor geometry and the pc economy, a megabyte of memory has gone down below U$0.1 in Q3, 2001. The price spiral has reversed in December of 2001. The new generation of Double Data Rate (DDR) memory has pulled the figure back up to U$0.27 / MB in the spring of 2002. Upward price pressure persists into Q2, 2002 as DRAM manufacturers are transitioning into the 256Mb generation of DDR DRAM chips.
The DOT.COM Bubble Burst
The Dot.Com downfall of 2001 created a sudden surplus of computing and communications equipment. Lot of servers, routers and workstations became warehouse inventories. High-tech companies had to let off people, which created a down economy and a recession. With no new killer applications with PC, Moore''s Law of PC growth is being challenged.
Note: Moore''s Law essentially states that the PC is replaced with new machine twice the performance every 18 months.
Intel Was Dragging Its Feet
While the Rambus vs Infineon SDRAM/DDR infringement case was pending in court, Intel hesitated to change its position on Rambus support. At the same time, Rambus sentiment failed to bring the price of Rambus memory down to the consumer acceptable level. OEM computer manufacturers were crying out for Intel to embrace the new DDR memory. Micron declared DDR price parity mid-2001 and set the new memory adoption on its way. By May 2001, the court had the verdict. Rambus'''' court lost essentially nullified its ownership on SDRAM/DDR patents. Intel had no choice but to announce the 845D chipset that supports DDR memories.
Read Historical Article: A WAKE UP CALL FOR INTEL ?
DRAM Manufacturing Consolidations
The great memory depression of 2001 had probably changed the DRAM manufacturing landscape entirely. With the high debt ratio left down from the last fab upgrade, companies have difficult time to justify their next upgrade to the new 300mm fab. Consolidation is a must for survival.
Micron Buying Up Hynix
Perhaps the biggest news in the memory industry this year is Micron''''s offer to buy Hynix''s DRAM manufacturing plants for $4 billion dollars. Considering the current state of market, analysts think that it''s a good deal for both parties.
Hynix was hardest hit among top global memory makers by the collapse in chip prices last year, posting a net loss of $3.85 billion. The world''s third-largest memory chip maker and its creditors entered talks with second-ranked Micron in December, 2001. If the deal were struck, Micron would surpass top-ranked memory maker Samsung Electronics Co. A sale to Micron would offer Hynix''s close to 100 creditors part of the $6.6 billion they spent saving Hynix last year.
Since Hynix''s plants are mostly the older 8 inch technology, it would take Micron a period of time to upgrade and to integrate the product manufacturing process. If the merger materialized, we can see a short-term reduction in Hynix capacity during the change over.
Read Latest News: Hynix board rejects sale to Micron
Toshiba Sold Off Virginia Fab
In a historic decision for Japan''s semiconductor industry, Toshiba Corp. said on Dec. 18, 2001 that it will withdraw from the commodity DRAM business and sell its Dominion Semiconductor fabrication facility in Manassas, Va., to Micron Technology Inc. for an undisclosed sum.
The deal with Micron was struck just as Toshiba was reportedly nearing an agreement with Infineon Technologies AG involving its DRAM operations. But the deal with Micron will enable Toshiba to quickly and completely exit the DRAM business, said Toshiba president Tadashi Okamura. The deal was completed on April 22, 2002. Six-hundred of the facility''s 1,400 workers were cut as a result of the deal.
Prominent DRAM Manufacturers Acquiring Capacity in Taiwan
On March 18, 2002 Mitsubishi Electric Corp. has agreed to license its latest process technologies to Taiwan''s Powerchip Semiconductor Corp., underscoring the Japanese firm''s interest in the memory IC market.
To meet rising demand for memory chips, Infineon Technologies A.G. has turned to Taiwan''s Winbond Electronics Corp. and Mosel Vitelic Inc. to secure additional capacity. Infineon said that it has signed a non-binding memorandum of understanding to provide DRAM technologies to both companies in exchange for production capacity. At the same time, Nanya Technology Corp. is also in talks with Infineon Technologies AG regarding a potential collaboration. Infineon has decided to team up with Nanya to construct two 300mm wafer fabs to produce DRAMs in Taiwan.
Elpida is Changing Direction on Memory
Elpida Memory is the new brand name of the DRAM joint venture between NEC Corp. and Hitachi Ltd. The union was finalized in September 2000. Although Elpida has already developed its first 0.13-micron design-rule 256-megabit SDRAM, the deal has always been looked at as the two companies'' exit strategy from the bloody DRAM market.
The Great Silicon Rush in China
In September 2001, Semiconductor Manufacturing International Corp. (SMIC) has successfully run its first production wafer in the new Shanghai plant in China. In March 2002, it announced itself as the first semiconductor company in China to realize 0.18 micron production.
"China right now produces only a small fraction of the number of ICs domestically that are consumed in the Chinese market," said Daniel Wang, executive vice president of Shanghai=''s Grace Semiconductor Manufacturing Corp. (GSMC), which is slated to launch production by the end of 2002. "Customers increasingly will turn to foundries in China to meet their needs for local market designs, faster delivery, and lower price."
"It will take eight to 10 additional new fabs in China to even begin to meet the soaring demand here," Wang said.
China''''s foundries have also established close ties with the growing number of domestic and foreign-owned IC design companies that are setting up shop here (in Shanghai) SMIC''s Chang estimated there are more than 80 IC design houses in the Shanghai area alone. In fact, China''s foundries are hoping to take a page from their competitors in Taiwan, which developed a successful formula by servicing a local base of IC designers.
Already, foundries here have lined up large domestic customers and believe the burgeoning ranks of Chinese-owned and foreign OEMs and contract manufacturers will expand the demand for 8in.-wafer lines.
During the first week of April 2002 Taiwan government announced its more open policy on semiconductor investments in China. Two days later, the two largest manufacturers, TSMC and UMC announced their plans in raising fabs in China. The race is on to establish plants in China to meet the local demand.
Besides looking from the marketing side, we should also look from technological front. The evolution of technology and its process can sometimes stall the market progress in DRAM memory.
High Performance PC Design Requires New Thinking and New Tools
Although Intel''s Rambus design had previously wet the appetite of the design engineers, yet the DDR design revolutionized their thinking. Unlike the Rambus technology, which was totally a hand down from Intel, engineers found that they have to design new chipsets, new DIMM socket, new motherboard, and new memory testers. All these infrastructures had to be coordinated precisely. A coalition of DRAM manufacturers saw that and organized Advanced Memory International (AMII) to promote and to coordinate the DDR enablement. Due to the high frequency design, engineers had to learn circuit simulation, signal integrity analysis, impedance matching and new component selections. These had taken time for the change.
DDR333 Puts Confusion onto The Industry
Shortly after the first DDR266 DIMM was validated, Micron engineers discovered that a high percentage of their DDR production would meet 333 MHz performance. A push towards DDR333 was thus started. End users were confused to the fact if they should buy the DDR266 computer now or put off to wait for the delivery of a DDR333 enabled computer.
It took six months time for the first chipset company to declare 333MHz support. Silicon Integrated System (SIS) found out that its chipset designed for DDR266 can be over-clocked to work with DDR333. It had, therefore, placed their marketing around DDR333. For a while, it had given the end users some new hopes for added performance. The hopes soon fated when users found out they also have to over-clock their CPU to the unreliable zone to accomplish the 333MHz memory access.
Although VIA Technologies later announced the KT333 chipset and the V4X333 chipset that would work with DDR333 without over-clocking the CPU, it never carved out a significant piece of the market. On the other hand, Intel insisted on to first stabilize the DDR266 market before any new introduction of DDR333.
However, the most important factor concerns JEDEC, the memory standard forming committee was that some tweaking have to be done on their PC2100 (DDR266) Gerber to achieve robust PC2700 (DDR333) performance. These changes had, unfortunately, delayed the adoption of DDR333 for at least 3 months. By Q1 of 2002, JEDEC task groups were working vigorously to catch up their lost time.
With all the activities going on, the general consensus of the industry is still Q4, 2002 before mass adoption of DDR333 can happen.
Read Informative Article: What is PC2700 DDR SDRAM ?
Larger Wafer and Die Shrink Adds Difficulty
As new operating systems are demanding more memories, memory cost has to drop to maintain the cost ratio of memory in relationship to the entire computer. This cost reduction can only come from semiconductor die shrink and larger wafers. That means the industry had to wait for 0.18-micron process to produce an efficient DDR266 DRAM chip. Likewise, 0.15-micron process would be preferred to produce the DDR333 DRAM chip. DRAM manufacturers also have to migrate from 8-inch wafer to 12-inch wafer to compete in the market place.
At the same time, smaller geometry also requires better dielectric to construct the same capacitor with smaller physical size. That means new material science has to go hand-in-hand with the progress. Any glitches can result in delay of the memory industry.
Memory Hungry Servers Failed to Deliver Their Promises
As early as Q1 2001, servers were predicted to take up 40% of the memories in the entire market. Due to the Internet network slow-down, actual server memory usage was far less than predicted. This situation will likely linger on throughout the rest of 2002.
Is DDRII On The Horizon?
DDRII is supposed to be the next generation memory after the present DDR (DDR-I). JEDEC''s schedule is to complete the chip specifications by Q4 and have prototype DRAM available by Q1, 2003. Production realization and modules are scheduled for 2004.
The key features on DDR-II are:
Differential clock and differential data lines.
Post Cas operations for better latency.
On die programmable terminations for best impedance matching.
Off chip driver control to obtain best signal integrity performance.
Since the last two features were originally created by the Intel hosted ADT alliance and brought into JEDEC at the last stage, it will likely delay the confirmation of JEDEC DDR-II specifications. On the other hand, SSTL-1.8V would be best served with 0.1 micron or better geometry, it is not expected to reach price parity and mainstream adoption level until 2005.
Memory Event Outlooks
DocMemory see that DDR266 will roam for the rest of 2002. The older SDRAM will remain and taper down for the next two years. DDR will transition to 333MHz early 2003 when most of the manufacturers have completed their factory equipment upgrade. DDR-I 400 will appear and stay for boutique applications. DDR-II will likely come in the 2004-2005 time frame. It will start out at 533 MHz and quickly migrate to 667MHz.
New Application Needed to Revive PC Industry
As for new memory applications, the new eHome push will connect more home appliances with the home computer network. These connections will be in Bluetooth or 802.11. The connections will enable remote control of home appliance through the Internet. The smart phone will double up as a portable controller. To achieve higher performance, the smart phone and PDA will adopt DRAM as their main memory. DRAM memories will be used in more point-to-point constructions instead of DIMM format.
Prices! Prices! Prices!
As for DRAM prices, DRAM manufacturer consolidation will result in less number of players. Less players will mean price stability. The only remaining factor to keep DRAM price down would be consumer affordability. The weak economy might keep computer sales down and thus force on price cuts. We will see price level of DRAM being tested over and over again until the general economy can sustain some price inflation.
Production Landscape Will Change
With all the new 300mm fab being built in Taiwan, it will increase its position tremendously. We will see Taiwan occupying 20% of worldwide DRAM supply market. Japanese DRAM manufacturers, instead, will use Taiwan and China subcontractors for production. They will maintain their R&D superiority in DRAM design and innovations. China, the sleeping silicon giant, will wake up to be a major memory production center. They will repeat the path that Taiwan had gone through a decade ago.
Copyright © 2002 CST, Inc. All Rights Reserved