Although I’m not sure molecular biology is a must, having a class in it is not a bad thing, if only because biology is the only worked example we have of a fully functioning nanotechnological system (though one would hope that engineered systems would be far less messy and disorganized.)

Many budding chemists skimp on learning physics — don’t do that. Physics is pretty important too.

@ jim moore — You asked me to elaborate on what my remark that building a tiny bit of diamond would be an impressive lab demo, but that I don’t think its likely that the methods could be extended to produce results that are very useful in a general sense. What I have in mind is the relative difficulty and rewards of competing approaches, some of which have gotten far too little attention. I’ll be saying more about this, of course.

When looking down the road, it’s important to distinguish current, intermediate and advanced technologies and capabilities. Diamond and related materials are excellent, and that using them in nanostructures will eventually be practical for even large-scale applications. My remarks are directed to current methods, extensions of them, and intermediate developments

@ Markku Jantunen — Yes, larger memories shouldn’t be regarded as merely providing more space for existing software, but as an opportunity to restructure software to shift the memory/processor trade-off. A technique that I’ve used from time to time is memoization.

@ Phil Duby — As you say, FPGAs can be implemented with highly regular structures, and this makes them candidates for relatively early development. They’re a way to make complicated circuits from simple structures while avoiding complex fabrication. In comparison to memory, there would be a greater premium on speed, and the unit cells would of course be more complex. Ensuring tolerance for fabrication defects would also be more complex, but an effective approach is to locate the defects and then customize the mapping of the logic circuits to the physical structure, as demonstrated in the Teramac work [pdf] at HP. Their machine worked despite a 10% defect rate in its programmable logic cells.

Space and time can be traded in algorithmics to varying degrees. If it should be possible to use molecular manufacturing to mass-produce cheap memory circuits of stupendlously large capacity (and fast access/write speeds), wouldn’t it make sense to use them to build computers (at least for special purposes) whose processors are relatively inefficient but whose memories are extraordinarily large, particularly if on chip photonic communication between components becomes commercially available?

What about metals? In Nanosystems you said only the hardest stiffest strongest metals with the highest melting points are worth it for MNT, I tend to agree.
But is there room for improvement with metal alloys and nanotechnology?