Re: DRAM and Flash production
I got the following from the HEA website; although it was written last year, it directly addresses the issue of production being cycled between the two products (bolded text is my highlight):
Using a DRAM-Based Manufacturing Strategy to Stabilize Flash Memory Price and Availability
Steven Grossman
Senior Vice President and General Manager
Flash Memory Division
Flash memory makers have figuratively taken their customers on a roller coaster ride since late 1992 when the Flash market really exploded. It has gone from about $200 million to today's $2.5 billion, and over the next five years it is expected to jump to $10 billion. However, to date, the OEM customer has fallen victim to the ups and downs of shortage and oversupply market conditions. In some years, there has been plenty of capacity, short lead times, and price declines. Other times, we've experienced severe product shortages, long lead times and product allocations, and dramatic price increases.
The dynamics of Flash supply and demand, in short, have, over time, created some major business issues for OEMs. In 1992, as demand dramatically escalated, a major Flash supplier decided to transfer production from one U.S. fab to an overseas foundry. That transfer ran into major difficulties which, in turn, caused product shortages and industry price increases. The next year, demand growth was stunted and prices collapsed as new capacity was brought on line. In late 1994, the market exploded again and suddenly everybody wanted Flash memory. Product went on allocation; prices shot up again. These cycles have continued with Flash currently in overabundance and prices quite low. No one really knows what to expect in 1998 and 1999.
Obviously, this does not bode well for OEM customers who are, in most cases, trying to stay at the leading edge of their respective markets. Flash customers have suffered the adverse effects of this topsy-turvy market. They hate these fluctuations because while they're trying hard to sustain a massive business level in their dynamic markets, a relatively small scale component is creating havoc with their product design and production cycles. Smaller OEMs have even been forced out of the market because of limited Flash allocation. Ideally, to completely avoid these Flash availability conditions, Flash customers want a multi-sourced commodity product with stable pricing and availability.
The best solution may come with the availability of Flash built in high volume DRAM fabrication facilities. DRAM is the highest volume commodity semiconductor product, supporting an order of magnitude more fabrication capacity than Flash. Consequently, a large DRAM supplier would have the capacity available to smooth out the supply and demand imbalances seen in the Flash market.
It is important in this regard to note that 90 percent of the Flash memory fabrication process is similar to the DRAM process. The other 10 percent requires special fabrication know-how to cost-competitively manufacture high quality Flash memory in production volumes and with high reliability. A lot of this manufacturing skill and knowledge focuses on how to make the Flash memory cell uniformly and consistent so that the product can achieve the necessary high reliability and endurance. Hence, when Flash manufacturing can take a "free ride," so to speak, on a DRAM fabrication line with very minor modifications, a commodity product is the end result.
Some have questioned the feasibility of this. For example, early on when Flash first started appearing, the EPROM and EEPROM vendors making Flash products touted the idea that expertise in tunnel oxide was essential for manufacturing high yielding, high quality Flash chips. They claimed that only non-volatile memory suppliers would migrate to Flash. This may have been true five to ten years ago when Flash was in its infancy. But today, the "black magic" associated with Flash is widely available throughout the industry, including the large suppliers of DRAM.
In fact, expertise in tunnel oxide technology is not nearly sufficient to serve the needs of this market. On the contrary, today, you need to know how to make highly acceptable, volume production Flash products based on advanced deep submicron process technologies. And this is where having a high volume commodity memory process makes a major contribution to stable pricing and availability, as well as a big difference on who has and who doesn't have the right Flash product at the right time and at the right price.
Fabricating Flash products via a microprocessor process driver is another technology/market force that generally works against the idea of supplying the industry with a commodity product. Consider that Flash uses two to three layers of polysilicon and two layers of metal. On the other hand, a microprocessor process entails about five metal layers and one layer of polysilicon. In effect, microprocessor fabs aren't properly set up to run Flash memory because the number of metal and polysilicon steps are different. The equipment mix is all wrong to run both Flash and microprocessors.
Still, some vendors are attempting to run both on the same production line. But when you do that, there are countless inefficiencies and the probability is high you can increase the total cost of both sets of products if the mix of equipment isn't perfectly executed when the fab is designed and set up. When chipmakers like these do set up their hybrid Flash/microprocessor fabs, they've got to reach a proper balance of metal and polysilicon process equipment. Otherwise, they won't be able to cleanly and economically shift capacity to and from Flash and microprocessor production.
Conversely, if a high volume DRAM producer has the crucial and necessary technology know-how, that particular vendor can easily and economically shift capacity to Flash production in order to meet unexpected product demands. For example, at Hyundai Electronics, we have four state-of-the-art, eight-inch, 30,000 wafer per month fabs running 16- and 64-megabit (Mbit) DRAM. This constitutes enormous capacity.
In our case, if an OEM, regardless whether he's small or large, comes to us for an unexpected million-piece order, all we have to do is quickly convert some DRAM wafer starts to Flash wafer starts. A million-piece order of Flash is basically a drop in the bucket when you consider the astronomical DRAM production figures of today.
Critics say running Flash on a DRAM fabrication line is not that easy. My answer to that is, yes, it can be difficult. But when you have dozens of experienced Flash engineers dedicated to design and production of quality Flash products and a proven track record, doubts like this evaporate pretty quickly.
Again, the idea is to create a production environment that will produce a commodity Flash memory so that the OEM can have stabilized low prices and off-the-shelf availability. Flash will continue to be a burgeoning business. But think about how much better this business can be for all concerned if we can eliminate the historical supply/demand imbalances. It makes good sense to find ways to effectively supply OEMs with Flash product so they can, in turn, maintain more organized production schedules with predictable cost.
3101 North First Street, San Jose, CA 95134
Phone: 408-232-8000
Copyright © 1998 Hyundai Electronics America.
|
