This month’s IEEE Computer Magazine has several articles on “rethinking computers”. (This year is the fiftieth anniversary of the IEEE Computer Society, with which has been marked with plenty of looking back.)
In an interesting article Christof Teuscher considers some “blank slate” ideas for totally new approaches to computing.
He sketches some familiar problems for conventional technology, heat dissipation, energy consumption, physical size limits on transistors, and limits on clock speed. He also notes the prohibitive costs of further improvements.
Teuscher is a hardware guy, so he leaves off the fact that we barely know how to program the chips we have, and such software we do have is laughably insecure and unreliable.
So, as we are now trying to reboot computing , to find innovations that go beyond incremental improvements?
Teuscher suggests one source is to look at older ideas that have already gone through part of the “hype cycle”, such as neural networks. From high hopes and unreasonable expectations through disappointments, neural networks have arrived at “the plateau of productivity”. Actually, they are ubiquitous, though always in hybrid systems with mixed architectures.
“unconventional-computing technology seems to have trouble cross-ng the Trough of Disillusionment and reaching the Plateau of Productivity. “ (p. 54)
He proposes a catechism for proposed new technologies.
- What challenge (or problem or application) are you trying to address with an unconventional- computing approach?
- What are the metrics for meet- ing that challenge?
- How is the system controlled and programmed?
- What fundamental limits to computing should you be concerned about?
- How do you interface with your unconventional system?
This is as good of a list as I have ever seen.
What is this “Unconventional Computing” Teuscher looks for “The Weird, the SMall, The Uncrollable.“ He calls this “intrinsic” computation.
“By using the general concept of intrinsic computation,we can harness a substrate’s intrinsic dynamics to perform use- ful information processing—that is, solve a given computational task..” (p. 57 )
One general point is that CMOS chip design is top down, and dedicated to difficult (and expensive) manipulations of Silicon to make what we want. Bottom up approaches take the natural behavior and properties of a system to create a device. “the lack of control over bottom-up self-assembly processes often leads to designs we cannot fully control and understand.” (p. 55)
He describes the example “Reservoir Computing”, which was proposed by Alan Turing. These systems are bottom up and uncontrolled, but he says that the idea is actually quite reasonable for many new nanomaterials and processes.
Chemical and biochemical systems, including DNA, are another “Unconventional” approach. These systems are small and fast, and are programmed via learning. Scaling and hierarchical composition are still challenges to be solved.
Teuscher summarizes these examples as demonstrating how these are different from conventional computing machines.
“First, to perform useful computation, we do not have to be in complete control of the physical substrate. Quite the contrary.”
And second, “many nontraditional computing approaches are highly specialized. “ There will not be a general purpose CPU, but a hybrid of specific devices.
The latter point is certainly “weird” to conventional techies, and, as he points out, means that these new technologies will depend on the ability to create and train new devices to solve each newproblem. This will be a great frontier for software and CAD systems, to say the least.
- Christof Teuscher, The Weird, the Small, and the Uncontrollable: Redefining the Frontiers of Computing. Computer, 50 (8):52-58, 2017.