SoylentNews is people
SoylentNews
IBM Reveals 8-Bit Analog Chip With Phase-Change Memory
Today at the IEEE International Electron Devices Meeting in San Francisco, IBM reported a new 8-bit analog chip. But the true development was less about analog chips catching up to their digital peers and more a radical rethink of chip architecture: this chip is the first to perform 8-bit calculations right where information is stored.
In traditional von Neumann chip architecture, data constantly shuttles between memory and processing, which consumes valuable energy and time, [...] IBM's new analog chip is based on phase-change memory. The key ingredient is a material that can undergo phase changes in response to electrical current. Typically, these are alloys of germanium, tellurium, and antimony. In one phase, which is conductive, the atoms are lined up nicely. In the other phase, which doesn't conduct electricity, the atoms move around, heated locally by current, and become jumbled.
(Score: 4, Insightful) by MichaelDavidCrawford on Tuesday December 04 2018, @09:06AM (1 child)
-tion.
Tsutomu Shimomura's first job after he dropped out of Caltech was to design a cellular automaton for Los Alamos that would do supersonic fluid flow simulations. While this is an important kind of simulation for reentry vehicles, that it was for Los Alamos in particular led me to speculate that it was for ICBM reentry vehicles. But really I never asked and he never said.
I expect it would also be useful for hydrogen bomb simulations.
Los Alamos wasn't very specific other than it had the idea this would be a useful thing to do. The great thing about cellular automata is that if you can come up with one that performs a useful calculation, it's easy to perfect it in software, with the individual cells being easy to design in hardware.
These are known as Lattice Gas Automata. I think his paper can be found but The Tubes aren't doing it for me just now.
I asked him what his automaton would be used for: "It would be about as expensive as a Cray" - I expect ten million back then - "but it would solve just this one problem at a thousand times the speed of a Cray".
Everybody things Von Neumann Machines are Soe Kewal that they have a very difficult time imagining other architectures. Electrical Engineers and academic Computer Scientists know all manner of better ways to build computers, but it is exceedingly uncommon for production coders to have a clue about these things.
Consider that in 1969, Armstrong and Aldrin landed on the Moom with 32 KILOBYTES of Magnetic Core Memory. Thus you can see from the very start, everyone who had serious work to do on a computer knew very well that only concurrent processing could ever hope to serve their needs.
Consider that the very FIRST applications of true Turing complete computers were codebreaking and hydrogen bomb simulation. Back in those days, if you wanted to model "The Heavy", you had to crack out your soldering iron and some Radio Shack project books.
To design truly parallel algorithms is quite a different thing from monocomputer algorithms. Consider that you can't directly parallelize Quicksort, because the first pass requires a single-tasking pass over the entire file. Truly parallel sorts never need to do that.
Yes I Have No Bananas. [gofundme.com]
(Score: 2, Funny) by Anonymous Coward on Tuesday December 04 2018, @05:21PM
You misspelled "Mom". :-)