1.5 Machines and Computers on the Microscale and Nanoscale

౪

43

on this small scale, as always, surface effects predominate. That could lead to unintended

consequences, e. g., a starting material reacts with the silicon channel surface, the rough-

ness of the surface creates turbulent flow, hindering the reaction, or the resulting vis-

cosity stops all flow.

The analysis of any sensor signal is important; you must not only detect a signal, but

also know what the signal means. Looking at vision, for example, the only things eyes

detect are a large number of photons of different energy. Only after analysis takes place

in the brain, will you know that these photons mean that a lover is handing you a red

rose. That brain analysis is also the basis of intelligence and consciousness.

“Artificial Intelligence” (AI) is the attempt to mimic intelligence with computers. It

started out by optimizing the logic operations and algorithms to make each of the oper-

ations faster. This allows for quickly comparing different options and then choosing the

“best” option based on a set of parameters. Another AI method works with networks,

as the brain works with neural networks. A set of inputs acts on a lot of “neurons”, that

layer of operations then operates on the output layer of the network. These networks

can be optimized by “learning”. One common and established way to “machine learn” is

by pattern recognition (Figure 1.47). The computer basically “memorizes” the most com-

mon output patterns, and that information is fed into the intermediate network layer

via a feedback loop. The more operations that are performed, the more common outputs

identified, and thus the output becomes more accurate. This is how the computer Deep

Figure 1.47: Machine learning workflow.