Historical question, what went wrong with bubble memory?
|
From: John Mashey on 18 Oct 2006 23:09 JJ wrote: > kenney(a)cix.compulink.co.uk wrote: > > I can remember several years when magnetic bubble memory was > > going to be the next big thing, replacing most other forms of > > storage. It then seemed to disappear without trace. Has > > development stopped? > > > > Ken Young > > I also recall the jumps in sizes didn't increase very fast when the 1M > arrived and that DRAMs were starting to follow Moores law and could be > predicted out to follow scaling laws. Also they needed heaters to work > properly and permanent magnets to actually sustain the bubbles. > > Anyway the wiki bubble memory article brings back neural memories. Yes, and in fact, it would probably have been easier to have typed: bubble memory to Google, Yahoo, MSN, ask.com, all of which give that article as the first reference: http://en.wikipedia.org/wiki/Bubble_memory That is at least a reasonable article, although doesn't emphasize DRAM as one of the two technoloqies that squeezed bubbles. It correctly identifies Andrew Bobeck of Bell Labs as the prime mover in this; IBM (and others) also did a lot of work, and I haven't tried to count patents, but Bobeck was a key guy. Try: http://patft.uspto.gov/netahtml/PTO/search-bool.html and put in bobeck and bubble in the two search fields: many of the results are authored by Bobeck or references to his work... and that doesn't include his pre-1976 ones, of which there are a bunch. We spent plenty of money (I was at BTL 1973-83) on bubble memories, among other things because they were non-volatile and reliable, and couldn't afford to have disks. Also, these things did not generally fairl for reliability reasons, but got squeezed from both sides in the price/performance/speed hole between DRAM and disk, and classic graveyard of stroage technologies. Moral: one must always be careful wth information found on the Web, but in this case, the most trivial search retrieved a fairly good exposition. When you ask a question in a newsgroup, you get answers that: a) Are now archived "forever" b) and are ... wrong :-)
From: kenney on 19 Oct 2006 09:30 In article <1161227370.927487.258010(a)i3g2000cwc.googlegroups.com>, old_systems_guy(a)yahoo.com (John Mashey) wrote: > to Google, Yahoo, MSN, ask.com, all of which give that article > as the first reference: I don't have broadband and pay for connection time. I use an off line reader for Usenet. Thanks for the references. Ken Young
From: Eric P. on 19 Oct 2006 10:03 kenney(a)cix.compulink.co.uk wrote: > > In article > <1161227370.927487.258010(a)i3g2000cwc.googlegroups.com>, > old_systems_guy(a)yahoo.com (John Mashey) wrote: > > > to Google, Yahoo, MSN, ask.com, all of which give that article > > as the first reference: > > I don't have broadband and pay for connection time. I use an off > line reader for Usenet. Thanks for the references. > > Ken Young Ok, then go straight to the horses mouth. A Y2000 interview with Bobeck in which he states at the end the reasons for its demise: cheaper semiconductor ram and disks. Bubbles: the better memory http://www.eet.com/special/special_issues/millennium/milestones/bobeck.html But research has continued, at IBM at least. In Sep, 2006 Nature Oscillatory dependence of current-driven magnetic domain wall motion on current pulse length http://www.nature.com/nature/journal/v443/n7108/abs/nature05093.html appears to deal with moving magnetic domain walls around using electric current. Eric
From: Eric P. on 19 Oct 2006 11:23 Tim McCaffrey wrote: > > I use to have a databook on these, probably threw it out... > > Anyway, one problem I recall with BM was that the bubbles were > arranged on rings, there was one primary ring hooked to multiple > secondary rings. A bubble was rotated from a secondary ring to the > primary ring, and then moved under the read "head" (this is all from > memory...), which did a destructive read, and wrote it back. You > had to be careful to rotate the rings to a known position before > power off, so the memory was non-volatile, but you could scramble it > easily enough. > > - Tim Oh yeah, that sounds right. You just tweaked a neuron. I remember thinking that you will need to devote one secondary ring to storing a single bit to mark the origin position. Also some of the secondary rings were bad so you had to skip those positions on the primary loop when storing/retrieving. But I see from poking about the web that later Intel added an extra secondary 'boot loop' to store both the origin marker and good/bad map which their controller chip would load on start up. Also a price of $100 each (in 1977 dollars) pops to mind. Intel was apparently using synchrotron Xray lithography (yikes!) to get the 1.2 um features for these puppies, so maybe that is why. I also have a vague recollection, which I have not been able to verify by poking about the web, of there being scaling limits to the technology such that they couldn't get much beyond 1e6 bits/cm^2 (something about a minimum bubble size and the bubbles can't be too close to each other or they merge). But I find no reference to that so caveat emptor. Eric
From: Christopher C. Stacy on 19 Oct 2006 21:50
"Derek Simmons" <dereks314(a)gmail.com> writes: > kenney(a)cix.compulink.co.uk wrote: >> I can remember several years when magnetic bubble memory was going >> to be the next big thing, replacing most other forms of storage. >> It then seemed to disappear without trace. Has development stopped? > > If I remember right it was very slow, very expensive, not completely > reliable and I think IBM held most if not all the patents. > Being slow might have been related to reliably reading > and writing to the device. It was used for secondary storage on a version of the TI Silent 700. I think the FTD (Florist) network was based on these. |