This is a short maybe heretic post on the topic of architecture exploration.
It just struck me the other day that the idea prevalent in chip design that you want to explore the design space of a certain with great detail and precision might be the wrong way to go about things. It is in some sense similar to the ideas of planned economies: you decide a priori what is important, and try to optimize the design to do just that. If you are right, it is brilliant. If you are wrong, it can be very wrong. That is scary to say the least. It seems to assume that you have a pretty good idea of what you need to achieve, and that this is not likely to change for the lifetime of a design.
In product design in general, this tends to be a brittle strategy. In many areas today, especially in consumer-oriented markets, the approach that appears the most successful is to throw things in there and see what works. It appears to be more successful as a strategy to launch ten different mobile phones that might well overlap in target markets and feature sets, and then let the market figure out what was the good and what was bad. In foods, instead of having a single flavor of a product, you launch a bunch and then discontinue the ones that did not sell too well. For example, some of my favorite cookies did this a couple of years ago, branching out from plain nougat to mint and banana and orange and toffee… the banana is apparently dead now, judging from the Ballerina homepage.
Anyway, I can see the reason why chip design is not done this way: the cost of each try is forbiddingly high… but maybe if we could find a way to make it cheaper to just try things? Could we then move back to a more relaxed creative environment where being wrong ain’t so bad as long as something else worked out?
We are doing that on a grand scale, really, in the way that different chip startups pop up with a truly new idea, and the market tells us what works and what does not. The kinds of architectures being explored that way are usually much more radical than what a traditional chip design house will come up with in a “architecture exploration” phase. The cost of failure for a chip startup is certainly non-neglible, but I hope something will happen to help bring the cost under control so we can continue to explore the truly new architectures by the mechanisms of the free market…
And by the way, I like the traditional Nougat the best, even if Mint Chocolate can be pretty OK too.