It's also possible to configure the switches to work as logic gates, so that they change state depending on a combination of inputs. Creating a circuit that detects when either or both of two inputs are present is a simple matter of arranging switches in the right order with some extra components: it's been far harder to make an inverter that's got an output of one when the input is zero and vice-versa. Called a NOT gate, this is essential for all computer logic -- and it is this that HP recently announced.
It's an involved process, turning a signal upside-down with just switches. In effect, a high signal sets a switch that couples a separate low signal to the output, whereas a low signal doesn't change a switch that's set to high, but to make this happen properly in a grid of switches which share their inputs and outputs a series of clock signals have to switch between components in the right sequence. A beneficial side effect of this is that the output signals are restored to full strength, just as long as the inputs are good enough to change the switches -- this regeneration process is essential in complex circuits with many stages.
At this point, you have all the building blocks needed to duplicate existing processor designs in the new technology. All that remains is making the new designs economic to produce and reliable enough to work commercially.
That is a huge task, and one that nobody active in the field underplays. There is considerable interaction between wires carrying different signals a couple of nanometres apart, and when you combine thousands or millions of such wires into a small space it can be nightmarish telling wanted signals from crosstalk. HP's approach is to subdivide the big grids of wires into isolated patches by introducing changes in the chemistry of the wires at the boundaries. Producing the circuits may be as simple as stamping them out using a version of contact printing: without the need to align tens of layers of careful chemistry, the potential is there for even greater economies in production than transistor-based silicon enjoys.
The first practical devices using these techniques will probably be hybrids, much as there was a 25-year period where transistors and valves coexisted, often in the same circuits. Work is underway to modify existing CMOS design, test and production processes with nanotechnology capabilities; a memory circuit might use existing transistor engineering for its interface to the outside world and to control a core array of roxatane storage devices. There are other chemicals that can be used in crossbar architectures too; they don't have the full potential range of benefits roxatanes enjoy, but may be more amenable in the short term. Still others may be better all round. Work continues.
There is no guarantee that any of these ideas will successfully see full production, but as the researchers and engineers conquer each new problem their confidence is growing. New ideas impracticable in silicon -- such as folding layers over on top of each other -- are helping to feed that excitement. Stan Williams, director of HP's Quantum Science Research group, is on record as saying "I think we've picked a winner -- something that will allow Moore's Law to continue for another half century. I used to think that was impossible. Now I think it is inevitable."
