Dataflow in molecular QCA: Logic can “sprint”, but the memory wall can still be a “hurdle”

As we approach the single molecule scale, we are fast approaching the the limits of miniaturizing information. The star of silicon microelectronics has been the field effect transistor (FET) which acts as a voltage-controlled current switch, and a bit is represented by the on/off states of a pair of n-channel and p-channel FETs. However, porting FETs to the molecular scale has its limitations. One limitation is the difficulty of connecting different molecular devices. Making one device is not sufficient. A signal in one molecule must switch a signal in another, and so on down a cascade. The FET is a three terminal device and current must transfer charge onto the gate to open or close the conducting channel. Constructing molecular wires that connect the outputs of each molecular transistor to the inputs of the next represents a considerable challenge. At this point, no demonstration of even an isolated molecular FET has been made. Molecular transistors will also dissipate too much energy as heat. Consider a hypothetical molecular FET with an average footprint a operating at frequency f . Assume that all of the problems of interconnection have somehow been solved, and the average footprint accounts for source, drain, and gate. Now, assume that in each clock cycle Ne electrons move from the power supply to ground through the transistor channel. Each electron transiting the channel dissipates energy eVdd. The dissipated power density is then P= f eNeVdd/awatts/cm 2. Even with a= 100nm2 (1012 devices/cm2), a supply voltage of Vdd = 0.1V, f = 100GHz, and Ne = 10, the power dissipation will still be 16kW/cm2. This analysis holds no matter how the transistor is made. In we want to continue scaling device sizes down and performance up, other ways to process and propagate information must be studied.