Faster Computing through Chip Flaws. One of the more interesting lines of research at chip manufacturers is to make chips better by making them perform worse. MIT has done research that shows that many of the tasks that we perform on computers such as looking at images or transmitting voice don’t require perfect accuracy. Yet chips are currently designed to pass on every bit of data for every task.
MIT has shown that introducing flaws into the data path for these kinds of functions can speed up computing time while also cutting power usage of a chip by as much as 19%. So the MIT researchers have developed a tool they call Chisel which helps chip designers figure out just how much error they can introduce to any given task. For example, the program will analyze the impact of making mistakes for 1% or 5% of pixels when transmitting pictures and will compare quality of the finished transmission with the power savings that comes from allowing transmission errors.
Computers that Don’t Forget. A few companies like Avalance, Crocus, Samsung and Toshiba make MRAM (magnetoresistive random-access memory) devices that are replacing older RAM and DRAM technologies in chips to provide non-volatile memory. The expectation in the industry is that these kind of storage devices will replace volatile memory (hard disks) within ten years because they are much faster and use far less energy.
There are a few initiatives working on improved MRAM technologies.NEC and Tohoku University of Sendai Japan has developed a 3D processor architecture where MRAM layers are combined with logic layers. The chip uses a technology they call Spin-Cam (content addressable memory) that promises to allow more fixed memory with faster access speeds.
KAIST, a public research university in Daejeon, South Korea has developed a chip they are calling TRAM (topologically switching RAM) that uses a phase-changing super capacitor to quickly write to non-volatile memory.
Computers with Common Sense. The Paul G. Allen Foundation is awarding grants to projects that aim to teach computers to understand what they see and read. The projects will look at several different fields of machine reasoning to try to understand diagrams, data visualizations, photographs and textbooks.
The grants are part of a larger $79.1 million initiative into artificial intelligence research. This new research fits well into other Allen initiatives in deep learning to allow computers to explain what’s happening in pictures or to classify large sections of text without human supervision.
Quantum Memory. Researchers at the University of Warsaw have developed a quantum memory that will allow the transmission of the results from quantum computers over distance. Quantum computers operate very differently than Boolean computers in that they deal with probabilities rather than number crunching. Until now there has never been an ability to transfer the results of a calculation of a quantum computer because the very act of reducing it to ones and zeros destroys the result. For example, transmission of quantum results had no way to deal with the normal laser amplifiers in a fiber optic network.
The quantum memory consists of a 1 inch by 4 inch glass tube that is coated with rubidium and filled with krypton gas. When hit with a series of three lasers the quantum information gets imprinted onto the rubidium atoms for a very short period of time of perhaps a few microseconds. But this is enough time for the data to be re-gathered and forwarded to the next quantum storage device.
Self-Healing Computers. With hacking and malware on the rise, a new line of defense will be to give our computers the ability to heal themselves. Today we use a very static defense system for our computers consisting of mostly firewalls and virus checking. But anything that slips past those static defenses can be deadly.
There is an initiative at the Department of Homeland Security which is funding the development of a more active defense system that not only detects problems but which automatically fights back. The first stage of this new active defense is being called continuous diagnostics and mitigation (CDM). The goal of CDM is to enable each device in the network to self-monitor itself for signs of having been hacked. The first CDM systems will activate malware software to try to immediately rid the machine of the invader.
The next step after CDM will be to form a network-wide active defense that will allow networks to provide feedback about threats identified from individual CDM computers. In this next step the whole network will help to fight back against a problem found on one machine in the network. The ultimate goal is to create self-healing computers that continually make sure that all systems and data are exactly as should be.