Andrew Bobeck was out to challenge the entrenched memories of the late 1960s, ferrite cores and drums. Bubble memories were nondestructive and nonvolatile. Ferrite cores were good, but there had to be a better memory, one that wasn't so big, one that might be faster and, in time, one that might be cheaper. It would probably be magnetic because almost every known memory and storage device was magnetic. Ferrite cores themselves were magnetic; you changed the orientation of magnetization to store a 1 or a 0. Magnetic drums and magnetic disks stored information on magnetic materialon the surface of a drum or on the surface of a disk.
In 1967, a Bell Labs group working on magnetic devices included Andrew Bobeck, who started by designing signal and pulse transformersfast magnetics. He had worked on ferrite - core memories and had designed the first ferrite-core memory to be driven by transistors rather than bulky, fragile, heat-emitting and short-lived vacuum tubes.
It was a challenge because ferrite core owned the random-access-memory market and there were lots of vendors, so prices were low. The field in 1967 included core specialists like Electronic Memories and Magnetics, EMC Memories, Ferroxcube and Telemeter Magnetics, as well as the major computer manufacturers. One of them, NCR, developed its own alternative to ferrite core, tiny ferrite-rod memories.
Bobeck worked on magnetic logic devices and another recording medium, magnetic wire. (Before the advent of personal magnetic-tape recorders for audio recording, people used magnetic-wire recorders, which were notable for their uncanny ability to twist and snarl wire.)
Along the way, Bobeck invented the Twistor memory, in which a magnetic wire was twisted to produce a helical magnetic path due to the magnetostriction effect. If you drove a current along the wire's length, the magnetization, and thus the data, would move along a helical path.
The Twistor evolved. Instead of using a twisted wire, Bobeck used a copper wire and wrapped a helix of magnetic tape around it. There was a bunch of these in parallel, encapsulated in Mylar sheets.
The Twistor was nonvolatile, as was core. If power was interrupted, memory would remain. A Twistor modification was nondestructive, as the core was not. With core, every time you read out data, you had to write it back. The Twistor modification was based on running copper tapes, 80 to 100 mils wide, across the copper wire wrapped in a mag-tape helix. You applied the word current through the copper tape.
The object was to make the memory nondestructive, which was done by laying over the array a thin sheet of magnetic material on a phenolic board, with all the magnetic material etched away except at every intersection of a copper-wire horizontal and a mag-tape vertical. In effect, you had a small magnet at every intersection. You could then read out a 0 or a 1 from each local cell, depending on the direction of magnetization of the small magnet near it. This form of Twistor memory, the card-changeable type, was employed in the first commercial electronic-switching office.
The magnetic wire in the case of the original Twistor had to be stress-sensitive using material like nickel or iron. When you twisted the wire, you created a helical stress on the magnetization. Because the magnetic material was continuous, there was interaction between one bit location and an adjacent bit location, so you could move information along the wire, effectively creating a shift register. You could move a few hundred bits.
Thin-film magnetic memories also made a thrust at replacing the ferrite core. Thin-film memory was used in the Sperry Rand 1107 in 1962. In time, Bobeck's group at Bell Labs researched a thin-film memory with a continuous anisotropic evaporated magnetic film as the storage medium. During the investigation of storage-cell interactions, mechanisms for shifting data within the plane of the film evolved and a two-dimensional shift register was born. Unfortunately, the arrangement was crude; the mechanism for shifting data up and down was different from that for moving data left and right and movement was slow.
Trying to solve that puzzle, Bobeck had a thought. What kind of material, he wondered, would allow propagation in orthogonal directions based on the same effect. He decided that the required material should have an easy direction of magnetization perpendicular to the surface. That would be the only way for the symmetry to be correct. He started with manganese bismuth, which was extremely hard to magnetize; it was almost a permanent-magnet material.
Soon the magneticians suggested orthoferrite. Bobeck placed a platelet of orthoferrite against the surface of a block of ferrite channeled like a waffle iron. By running wires through the channels in the X and Y directions, Bobeck could magnetize small spots perpendicular to the surface and could move magnetic spots from one waffle-iron post to another post in either X or Y direction. And lo, he had the beginning of a bubble memory.
Islands of magnets
Orthoferrite had an easy magnetization axis, a not-so-easy axis and a really difficult axis. The magnetization would always lie in the easy direction. Then Bobeck noted that if you applied an external magnetic field to an area, you caused the island of magnetization to become perfectly round and very stable. All the magnetization was normal to the surface and that was the bubble. The bubbles were like little magnets floating on a liquid. But the orthoferrites didn't allow for small bubbles under, say, 1 mil in diameter.
Further work in materials led to garnet, believed to have no preferential direction of magnetization since the garnet had three-way (cubic) symmetry. Bobeck finally convinced others that garnet was made isotropic by its growth habits. And the garnet composition could be tailored to provide the desired bubble size. Further, garnet film could be deposited directly on a nonmagnetic garnet substrate.
In time, Bobeck's group created arrays that could hold 4,096 bits in a square inch, a tremendous improvement over the square foot required for an equivalent ferrite-core memory. In fact, the arrays eventually reached density of a million bits per square inch. Bobeck felt that it would be extremely difficult for core memories to achieve such a density because it was necessary to thread each core by hand. And a large bubble memory would probably be much less expensive than core.
Bobeck's bubble memories could be configured as random-access memories, but the emphasis was on serial memories rather than RAMs. Serial memories had about 10 times the density of RAMs. The target market was that served by disk memories and drum memories, and bubbles had several strong advantages.
Access times were shorter. And, more important for many applications, bubble memories didn't suffer from gyroscopic properties that posed drastic limitations on drum and disk memories. These gyro properties limited a vehicle's maneuvering speed, a significant problem in military applications. And bubbles could tolerate high shock and vibration environments, so they could be used near an atomic explosion or earthquake, as well as in military vehicles that might have to tolerate tough environments.
There was still another advantage to bubbles: They didn't wear out, while you never knew when a disk or drum would fail because of wear and tear. And there was another, more subtle, advantage, which was of great interest to the military. It was possible to wipe out secret information quickly and completely with an overall magnetic field. There would be no residuals like those a dedicated engineer could often recover in disk drives or drum memories.
Bubble memories captured the attention of the worldwide engineering community. They replaced small tape systems, eliminating the problems of tape wear and mechanical-system fallibility. AT&T used bubble memories to store recorded messages like, "You have reached a nonworking number."
Bubble memories were obviously the memory of the future. But there were some roadblocks. As bubble density increased, more and more difficulties were encountered in making the bubble paths completely error free. Material impurities or any other kind of imperfection would break a path in the memory and data would be lost. This, says Bobeck, was the Achilles heel of the bubble memory.
What burst the bubble?
That was a problem that might have been solved. So what, then, killed the bubble memory? Several things, Bobeck reports.
There was the tyranny of numbers. They were serial memories, so testing time on large bubble memories became significant. It took a long timeand therefore it was expensiveto test a large chip under different drive and environmental conditions.
Further, you can write software that allows you to skip a bad sector in a disk drive. The software notes that you have a bad sector so you simply don't use that sector. It's possible, also, to skip bad loops in a bubble memory, but when you initialize the memory, you have bubbles associated with those loops, so it's hard to get around them.
Probably most important, semiconductor memories came along and got better and better, larger and larger and cheaper and cheaper, as did magnetic-disk memories.
What happened to Bobeck? He switched to a study of laser diodes. In 1989 he retired from AT&T's Bell Labs with a record of more than 120 U.S. patents, more than anybody else active in AT&T. Then he switched again into watercolor painting and three-dimensional digital photography.
The Century of the Engineer: Misunderstood Milestones