December 6, 2000 9:01 PM PST
IBM, Infineon looking to shake up memory market
- Related Stories
IBM taps new technology to build faster chipsMay 22, 2000
Intel, memory makers form group to determine future chipsJanuary 17, 2000
The two companies, which will announce the effort Thursday, hope to speed development of Magnetic Random Access Memory (MRAM) chips. MRAM uses a magnetic charge--similar to a hard drive's--to store information, as opposed to the electric charge of existing memory forms.
MRAM has several potential advantages: It could be cheaper than standard computer memory, known as DRAM, as well as faster than the speediest static memory. In addition, like the flash memory used in cell phones and digital cameras, MRAM will not lose the information stored on it when it is removed from a power source.
As a result, MRAM could be used in potentially every sort of electronic device and eliminate any of the trade-offs that come when designers must select among the current types of memory.
"The key word is potential," said Bijan Davari, an IBM fellow and vice president of technology and emerging products. "In principal, it could do all of these things."
The challenge now is to see whether MRAM can be reduced to a size at which it will be cheaper and faster than conventional memory.
IBM, which has done most of the research behind the new memory chips, has already built sample chips to prove the technology works.
While others such as Hewlett-Packardand Motorola have discussed the potential of the technology, IBM said it holds an advantage, given its experience in the development of hard disks, which also use magnets to store information. MRAM takes a compound similar to those used in hard drives and implants it on the same piece of silicon as a traditional chip.
Huge market potential
If IBM and Infineon succeed, they could position themselves to grab part of a huge market for memory chips. Research firm Dataquest predicts that sales of standard DRAM alone will reach $30 billion this year, with the industry
Meta Group says market demand for memory for a huge number of consumer and business devices is growing at exponential rates.
The two companies could conceivably manufacture chips and license the technology to other chipmakers.
IBM and Infineon will work to take the technology from the research lab into high-volume production. By the end of next year or early 2002, it should be clear whether the technology can be scaled down enough to make it feasible and whether the magnetic material can be worked into traditional chipmaking processes, Davari said.
If all goes well, Big Blue believes the technology could be in commercial use by 2004.
Dataquest analyst Jim Handy largely agreed with Davari's assessment.
"It's a technology that I was at first a bit skeptical of," Handy said. "Seeing IBM get involved changes my perspective a bit."
A long road to improvement
Improving memory technology has been a problem for the industry for years. While processors have doubled in power every two years, memory speed has improved only incrementally. Relief has come recently in the form of two new and competing types of high-speed computer memory: Rambus RDRAM and Double Data Rate (DDR) DRAM.
Here is how the various types of memory stack up:
Dynamic Random Access Memory (DRAM)
Disadvantages: Comparatively slow, loses information when power is off.
Used in: PCs, game consoles and other electronic devices
Static Random Access Memory (SRAM)
Disadvantages: Costs up to four times as much as DRAM, loses information when power goes off.
Used in: cache memory
Advantages: Retains data when power goes off, low power consumption in read mode.
Disadvantages: Expensive, saving data is slow and uses a lot of power.
Used in: digital cameras, cell phones, MP3 players.
Magnetic Random Access Memory (MRAM)
Advantages: Potentially cheaper than DRAM, faster than SRAM, and can retain memory like flash when power is off.
Disadvantages: Won't be ready until at least 2004. Could be tough to integrate with other silicon-based chips.
Uses: Theoretically appropriate for all products.
Handy said that while MRAM may start to appear on a limited basis by 2004, it will likely be at least five years beyond that before it represents a serious threat to the various silicon-based memory chips.
While MRAM sounds great in the abstract, the companies will face challenges when it comes to manufacturing, Handy says. He cites the example of ferroelectrics, a technology that has been touted for 20 years as an alternative to traditional silicon-based chips but has never made a serious dent in the market.
"We've seen a lot of technologies that have held promise," Handy said. However, none have matched silicon in terms of cost and ability to be integrated with other parts needed in computers and other electronic devices.
But if IBM is able to get the costs down, Handy said, MRAM is "something that the industry would sit up and take notice of."
That's because as chips get smaller, the individual circuits hold less of a charge, making the chips harder to manufacture and increasing the risks of leaking current and other problems. Nonetheless, Davari said he expects some form of today's silicon memory to survive to battle against MRAM even if it does make it to market.
IBM has been looking to beef up its chipmaking operations. The company said in October that it would spend $5 billion to build a new chipmaking facility and expand existing plants to support its growing semiconductor business.