Then and Now in Memory
Monday, October 28, 2002
DRAM - the "black gold"
In early 1996, I attended the Cebit [the largest computer & electronics trade show in the world] held yearly in Hanover, Germany for the first time. Surprised by its sheer side of exhibition was out of question. But one thing that struck me the most was the DRAM module buying experience. Memory modules exhibitors were everywhere among the 20 over exhibit halls at the show. I could not help but running into them every corner I turned. Let me clarify, I was there as a rookie exhibitor and not the buyer. But I ended up buying 4 pieces of 32 Megabytes (MB) DRAM modules at the price of US$190 each from one of the countless memory module exhibitors.
The market price for a 32MB DRAM module at the time was around $500 each. The deal was "too-good-to-pass-by" that I begged my manager to loan me some travelling money. After some convincing persuasion, I borrowed enough cash (cash was the only acceptable media in the transaction) to buy 4 pieces. I was overjoyed over the fact that I had just saved about $1,200 in one deal. I later installed a pair in my ex-girlfriend's PC (what an idiot) and sold the other pair for a higher price. Everyone including me, was finding ways to get involved in this so called "black-gold" rush. Even the robbers were targeting and putting their hands on the substance that was once more valuable than gold. However, anything that is "too-good-to-be-true" won't last long.
The good old times
Back in 1995, the memory prices shot up to the roof that anyone or any company in the distribution link wind up making tons of money out of it. Demand outstripped supply by a wide margin and only a handful of companies have the technologies and capacities to crank up DRAM in volumes to fill the increasing market demand that created by the Internet boom. The short supply made every byte and bit of the DRAM output valuable. Even the rejected parts were salvaged and turned into useful and money-making product for many "downgrade" memory module manufacturers.
Before long, the lucrative memory market started to attract more newcomers, many of those backed by respective government with unlimited power and resources, to create growing capacity. Existing suppliers continued to upgrade its process to improve yield and increase output. On the other hand, they also unintentionally helped remove the technological entry barrier by licensing the very process technologies to any new player willing to pay the price to enter the field. The whole market suddenly became too crowded. Every company produced the same common part and hence price was the only main competing criteria. The supply kept on escalating while the demand couldn't keep up the pace. The prices started to come down and come down fast, as if it were jumping off the plane with no parachute.
The saying of "what goes up must come down" always hold true. The memory cyclical began to take a turn after the peak in 1995. As the DRAM market continues its downward spiral, companies began to find ways to fight back or face the fate of being eliminated. DRAM suppliers started to diversify or expand their product mixes. Numerous consolidations took place since then among DRAM suppliers as well as module manufacturers to remain competitive. Countless so called "off-spec" module manufacturers shut their doors because the DRAM prices were so low that there was basically no price advantage of salvaging the rejected or off specification parts. Nevertheless, regardless of all the approaches employed to fight the sluggish market, the much anticipate DRAM price recovery still seems to be a distance away. Part of it is due to flat demand and the rest is caused by continue oversupply.
Circling the loop
To battle the weak market condition with plummeting prices, DRAM suppliers have to do something to reduce their costs and to remain competitive. To make the matter worse, the current DRAM prices have fallen so low that they are equal to if not less than manufacturing costs. DRAM suppliers are taking extreme measures to stay afloat. The only problem is that the very steps that they have taken are coming back around to haunt them.
One way to reduce cost is to shrink die sizes. Economy of scales suggests that more output translates into less cost per unit. Hence the efforts of reducing die size means more capacity and better cost management. For instance, the recent announcement of Samsung's leap to a 0.12µ process for high-end products, bypassing 0.13µ, is a good indication of the market leaders' strategy. Other example includes Micron Technology's plan to move from 0.17 µ to 0.15µ for its main products and then to 0.13µ next year. Infineon and ProMos on the other hand, will convert all the process to 0.14µ by 2002. Hynix and Elpida Memory are also migrating their die sizes from 0.17µ to 0.15µ as reported by their companies respectively. All these translate into double or triple capacity in the near future.
Another cost reduction strategy is increasing wafer size to 300mm. DRAM suppliers aims to gain more cost advantage through better and efficient technology. However, this requires taking on greater risks and higher investment than shrinking die size. Some suppliers are skeptical to invest in the new and unproven 300mm equipment. While others such as Infineon and Elpida see it as a way out and are planning to ramp up production in the third quarter of 2002 with the new 300mm wafer if nothing changes.
Despite what the economists say is a clear sign of oversupply in the DRAM market, almost all the major DRAM suppliers have maintained or increased production output during the current downturn. Therefore, as the economists put it, the DRAM market's current sluggishness is mainly due to structural weaknesses rather than just another cyclical downturn. Unless the established companies come to a consensus to stop increase manufacturing capacity and finding other ways to minimize costs, the oversupply issue is unlikely to go away in the near term.
The only other safe bet is the non-commodity DRAM for specialty markets. Even though such non-commodity represents only a small slice of the total DRAM market, it generates a large part of the profit. "One product that takes one percent of your production and sells with premiums offsets a lot of product that you sell at cost or below," said market analyst Bert McComas. To fight off the poor market condition, some companies diverts into manufacturing and marketing premium parts hoping to make up the lose margins. Samsung's recent release of its network DRAM is a good example. The Seoul, South Korea-based company has licensed IP from Fujitsu to produce the high performance DRAM. This network DRAM indeed is the same technology as Toshiba and Fujitsu's fast cycle RAM (FCRAM). The difference is Samsung network-DRAM is promoted to take the 10-Gigabit Ethernet and OC-192 optical network markets while FCRAM has been targeted at mobile applications. The other competing product that will soon come into the market is reduced latency DRAM (RLDRAM) advocated by Infineon Technology and Micron Technology. It is becoming a bit crowded even in non-commodity space. But niche markets will remain lucrative and companies will continues to find ways to create niche markets for their own products and as a way to combat the market downturn.
Like it or not, the good old days for DRAM were a thing of the past. As telecom and PC industries continue to weaken, the overall DRAM demand will remain in the slump. While some companies have the financial stability to hold on to what they do best and hoping for the market rebounds, others might face the cruel reality of being eliminated. The free enterprise will eventually work itself out. In the end, the companies with better management, strategy and resources will be the remaining few to catch the next rising waves.
Copyright © 2019 CST, Inc. All Rights Reserved