Do you remember Video CD (VCD)? It was an attempt by the multimedia industry to replace both VHS and laser disc, back in the early 1990s. VCD was limited to the capacity of a compact disc, meaning it stored 650MB and used MPEG-1 compression and stereo audio.
VCD wasn't very good. Video quality was subpar, as was audio. The technology didn't last long and it never took hold in the U.S., though it gained some ground in Asia. The major consumer electronics players behind VCD quickly regrouped around a far more advanced technology, DVD — and that was the end of VCD.
It's a common story in the tech industry: The idea lives on even if the specific technology does not.
Such is the case with bubble memory. That name probably means nothing to those of you not qualified for an AARP card, but back in the 1970s and 80s, bubble memory was the next big thing.
These days, you know the concept as flash memory.
In the Beginning
Like pretty much every other technology invented in the 1970s, it seems, Magnetic Bubble Memory came out of Bell Labs. The project was driven by a celebrated Bell Labs engineer named Andrew Bobeck, who helped design solid-state computers and magnetic memory devices.
image credit: Devlin M. Gualtieri
AT&T used bubble memory for its own devices, but the company also licensed it. Among its licensees was the premier memory manufacturer of the day: Intel. Before it became the CPU powerhouse, Intel's income came primarily from the manufacture of DRAM and SRAM memory.
"The whole purpose of all these new technologies, from bubble memory and leading up to today, is to find a new kind of data storage mechanism that is fast enough and dense enough and cheap enough to store all the data we're creating," said Bob Merritt, principal analyst with Convergent Semiconductors, who follows the memory industry.
Bubble memory was marketed as the next big thing for mass storage. Based on industry promises of the time, bubble memory would replace hard disks, which were monstrous, heavy, and had limited capacity for their size. Like the NAND flash of today, bubble memory would not lose its contents after power was turned off.
Its extremely low uncorrected error rate and overall ruggedness was especially attractive to the military, which saw it an ideal solution in hostile environments, said Robert De Cesaris, director of product development and manufacturing at Intel's Chipset and SOC IP Group.
Bubble memory stored data one bit at a time, just like NAND flash does today, in bubble-like magnetic regions on the surface of a chip – except the capacity was well below today's NAND memory. It was considered revolutionary when Bobeck's team showed a 1 cm square of memory that stored 4,096 bits of data. DRAM chips today are that size and hold up to 1Gbit of data.
By the mid-1970s, practically every large electronics company had teams working on bubble memory. By the early 1980s, almost all work on it stopped. Why?
So what happened? First of all, it turned out to be harder to make bubble memory than expected. The fabrication process never proved to be smooth or cheap enough to compete with other technologies.
"It's all about cost. Whatever is the lowest cost technology, no matter how flea bitten or ugly it is, is the one that takes over the market," said Jim Handy, principal analyst with Objective Analysis.
Plus, fabrication procedures in the 1980s were a little casual, to put it politely. Several companies, including Intel, polluted the ground water in the San Jose area. One Intel fabrication plant became a Superfund cleanup site, and other firms, like Fairchild and Intersil (which no longer exists) also had to clean up ground pollution.
The companies used trichloroethane, trichlorofluoromethane, dichloroethylene, and trichloroethylene, all powerful chlorinated solvents used to clean electronics parts. Trichloroethylene was also used for decades in the New England jewelry industry and other metallurgical industries to clean equipment and finished products, before the EPA decided, in the 1970s, that this stuff might be a little unhealthy.
"Trichloroethylene is very good of getting out of what you put it into,” said Handy. “It leaks out one way or another. So even though everyone was using it, they didn't realize how much would escape or the consequence of it escaping.”
One leak under Intersil wound up making its way through an underground well that traveled into the area of the offices of Siemens, and Siemens ended up having to clean up the mess. Intel Magnetics, the group making bubble memory, also learned what an escape artist these chemicals are, and it had to clean up a well in the San Jose area that supplied water to more than 300,000 people.
Bubble memory had other problems besides being messy to make. It required a complex controller similar to a hard disk controller, and it was power hungry and slow.
The fabrication process was unlike any other: There was no silicon anywhere. The final device required a host of exotic and costly materials: gadolinium gallium garnet (GGG) wafers, rare earth implants, and bias magnets to create stable bubble memory domains (the bits representing 1s and 0s), and ferromagnetic resistors and inductive coils around the die to move and control them, according to De Cesaris.
"The fabrication technology never proved itself to be smooth or inexpensive enough to compete with silicon,” De Cesaris said. “As the price of DRAM and hard drives dropped, bubble memories became relegated to a niche military market and were gradually squeezed out of existence.”
There were other issues, in addition to cost and manufacturing. In the early 1970s, IBM introduced a new hard drive, the 3340, more commonly known as the Winchester drive. Winchester drives were the first drives where the disk heads were not withdrawn completely from the stack of disk platters when the drive was powered down. Instead, the heads sat on a special area of the disk surface. This saved money, minimized complexity, and required less time to warm up.
DRAM and hard disk technology continued to accelerate in advancement, capacity, and reduced cost, which bubble memory couldn't match. By the early 1980s all efforts around it ceased.
"The problem for bubble memory was to make prices cheaper than hard drives, but since hard drives beat them at that game, then bubble memory didn’t have a leg to stand on. If bubble memory had gotten down to a cheaper price per bit, then people would have lived with it consuming larger amounts of space than a hard drive, so long as it wasn't orders of magnitude larger," said Handy.
Now, class, what can we learn from this?
Merritt recently attended the Flash Memory Summit and noticed that memory is undergoing some significant changes. He thinks DRAM is on its way out. "The old structure of DRAM and SRAM are breaking down,” he said. “We've gone on to so many different architectures, and new memory technologies coming. What it shows is OEMs are aggressively embracing the idea of cost performance tiers within their architecture relative to the memory.”
NAND flash and other charged storage technologies have serious long-term challenges. One of its biggest problems is the fact that the ability to store information is based on atoms and electrical charges in each cell. As you make those cells smaller, the probability of error increases.
That's why NAND flash makers are having a devil of a time getting their flash chips below 20nm. "The cost of maintaining Moore's Law for NAND becomes prohibitively more expensive," said Merritt.
But lots of candidates are ready to pick up the slack. ReRAM, for example, promises the storage capability of flash, the speed of DRAM, and the ability to shrink down to 4nm.
With ReRAM and other technical candidates popping up to challenge DRAM's dominance, Merritt noted memory OEMs are very aggressively looking for and are showing an ability to embrace new technologies when they show up.
And bubble memory, he added, was nowhere to be seen.