Supercomputers in your Pocket

  • June 7, 2018

    Supercomputers in your Pocket

    IoT brings connectivity to the internet to physical objects. This is a game changer for convenience, business intelligence and safety in some cases. Artificial intelligence (AI) can elevate these devices even further.

    A recent article published in MIT News reports on how mechanical engineers are leveraging AI and machine learning to enhance products we use in our everyday lives.

    Replicating the human brain’s ability to learn in a computer is incredibly difficult. First of all, both hardware and software need to work perfectly in sync. In the case of IoT devices, hardware features often include sensors, cameras, and light detection.

    Moore’s Law and AI for IoT

    MIT mechanical engineering faculty member, Jeehwan Kim, is working with his team to build new software specifically for intelligent devices. Moore’s law suggests the number of transistors per square inch on integrated circuits doubled each year since their invention. In other words, as time goes on, we can package more computing power into smaller and smaller devices. This is paramount for incorporating AI into IoT. Kim explains, “We are trying to build an actual physical neural network on a letter paper size.” The result? Kim calls it a “brain- on- a chip”.

    Mimicking the Human Brain

    The so-called, “brain- on- a- chip” replicates the neurons and synapses of a brain. Signals travel across thousands of cross points and arrays. Kim stresses this causes challenges as it is key to maintain control over the brain-like structures. Despite some of the formidable engineering challenges that lie ahead, Kim remains optimistic that soon we will be able to harness the power of a supercomputer in our portable devices.

    Improving Supercomputing

    Kim’s research group aims to address some of the limitations that plague traditional supercomputing. In its current form, supercomputing architecture requires data buses that cannot necessarily support real-time AI. The group is working on revamping ReRAM (resistive random- access memory) devices that can help overcome these limitations.

    With supercomputing in our pockets, the potential for our IoT devices is limitless.