The first practical example of a new kind of computer can complete calculations in minutes that would take a standard version longer than the age of the universe. This so-called digital memcomputer, which combines data storage and processing into a single component, could be a solution to the slowing pace of progress in traditional computer chips.
Theoretically, ordinary computers can solve any computational problem by carrying out single logical operations in a processor, storing the result in memory, and moving on to the next operation until the calculation is complete. In practice, this shuttling of data back-and-forth between components means some calculations take an impractically long time.
Memcomputers offer an alternative approach, using a single type of component called computational memory that handles both processing and memory functions at once. Many of these can operate in parallel, rapidly working together to find a solution. The downside is that memcomputers aren鈥檛 general-purpose problem-solvers, but must be designed to handle specific calculations.
Advertisement
Until now, those designs have been analogue, with computational memory units storing data as electrical signals of varying voltage. While these circuits were able to solve problems, tiny variations in these voltages mean that scaling them up also scaled up noise and quickly rendered outputs meaningless.
Now, at the University of California, San Diego and his colleagues have now constructed a digital memcomputer, using binary numbers to store data, that they believe can be scaled-up to handle practical tasks.
The team鈥檚 version is designed to solve a combinatorial problem common in standard computer chip design. 鈥淭hese are really tough problems,鈥� says Di Ventra.
A standard computer would need to methodically work through vast numbers of potential solutions sequentially until it stumbled across the correct one, but the memcomputer can find a solution in minutes. 鈥淚t would take over the age of the universe if you actually did it with standard algorithms,鈥� says Di Ventra. 鈥淭here鈥檚 no way you can do it.鈥�
鈥淪ince the 1960s, computational power has been largely driven by the ability to fit in more and more transistors,鈥� says at the University of Nottingham, UK, but chip designers are now reaching the limits of what is physically possible.
鈥淐omputing has become a little bit cleverer in terms of looking at new architectures and not being driven just by making transistors smaller and smaller,鈥� he says, adding that if research continues apace, then all computers could one day have dedicated memcomputing chips听to handle certain tasks.
鈥淚t鈥檚 under the radar a little bit at the moment, but we鈥檝e solved a lot of very tough problems,鈥� says Di Ventra. 鈥淲e鈥檙e interested only in really tough problems, because for simple ones you can always use standard algorithms.鈥�
arXiv
