[sdiy] Buchla 295 10-band comb filter topology

[brianw]

On Nov 25, 2022, at 4:51 PM, Pete Hartman <pete.hartman at gmail.com> wrote:
> On Fri, Nov 25, 2022 at 6:17 PM brianw <brianw at audiobanshee.com> wrote:
>> This was in the age when nobody had yet figured out how to stop cosmic rays from flipping bits and causing data errors at random. It used to be necessary to allocate this 12.5% extra in memory just to avoid using bad data without warning.
>> 
>> One of my EE professors told the story of one memory company discovering that the equivalent of Scotch tape backing material was enough to shield against interference from cosmic rays. They quietly added this to their DRAM chips, above the die, and watched everyone else go out of business. Proprietary knowledge, indeed!
> 
> I'm fairly certain that it must have been a little more complicated than that, because the Sun Microsystems, struggled pretty hard with this problem in L1 caches on their CPUs in the late 90's....  Until they put ECC on those memories too (having had it in their DIMMs for quite a while).  I had a rather ... boisterous customer who was hitting this problem, my boss sat in for me in a weekly meeting when I was out of town, and the next week I was gifted a roll of tin foil to block the cosmic rays because he had given them IBM's 1978 (or so?) IEEE paper on cosmic rays affecting memory with very little other explanation.

Now that you mention it, I recall that it has to do with the relative size of the silicon process versus the particle. When DRAM was less dense, cosmic rays had no (significant) effect. It was only when the density got high enough, and the DRAM cells small enough, that such a particle could upset the charge.

Given that this EE guest lecture occurred in the eighties, I wouldn't be surprised that smaller processes found problems with other particles or issues, or even that the shielding proven insufficient at greater densities. But one of the fun things about the story was that the metal enclosure of the computer was insufficient (as, presumably, tin foil would also be insufficient). I can't recall the name of the OEM that went out of business for want of an effective solution. As far as I recall, the story was meant as inspiration to think outside of the box and use multiple disciplines to find solutions, when necessary.

Do you happen to have a link to that IBM IEEE paper? I should probably clean the cobwebs out of my human memory storage.


It's certainly true that Flash memory fails for different reasons than DRAM, and there the ECC is used both to correct the error and signal that the data needs to be rewritten because the "nonvolatile" data has deteriorated and would benefit from a refresh before it becomes uncorrectable. It's also certainly true that modern DRAM modules are way more complex than simple SRAM chips.

Brian