Next Big Futures article The world is facing a computer problem: The bottleneck of computing power is our own minds.
The current form of computers, including smartphones, tablets, and the latest gaming consoles, use specialized chips for processing instructions and data.
However, the processors are limited by the way the human brain is wired, and it is not possible to create an entirely new processor.
Now researchers at MIT and the Massachusetts Institute of Technology (MIT) are developing a new type of processor that could theoretically be 10 times more powerful than existing chips.
They call this new processor “multimedia processor,” or MM.
The technology is being developed at MIT under a $250 million grant from the Advanced Research Projects Agency-Energy (ARPA-E) to develop a fusion processor that is more than 10,000 times more power efficient than existing chip technology.
The MM chip could be built using conventional, or “gate-driven,” semiconductor technology, as well as a “gateway” transistor that transfers electrical current between the silicon transistors.
MM chips could also be made from semiconductors that have high surface area, high die size, and high electrical conductivity.
The new processor could be used to run large, multi-core systems, such as a computer, or to provide new capabilities such as high-performance computing or deep learning.
MM can also be scaled up to scale a large computer system or be used in parallel to a single system.
The team has been working to develop MM chip technologies for more than a decade, using materials and processes that mimic the way silicon is designed.
“There is a big problem with the current silicon architecture,” says Dr. Anil Chopra, an associate professor in MIT’s Department of Electrical Engineering and Computer Science (EECS).
“There are certain things that are very difficult to do, and there are certain problems that are not easy to do.”
In recent years, MM has been explored by researchers around the world, including DARPA, which awarded MIT $100 million in fiscal year 2016.
The chip’s design is based on the idea that a processor is a “multichip” with more than one, sometimes two, transistors, which are connected by wires that connect them.
“We call them gate-driven, where the processor is connected to the processor by a pair of wires,” says Chopra.
“And we have to understand the logic of this circuit so we can understand what’s going on in the processor.”
Achieving the MM chip’s theoretical 10,200-fold power efficiency is a challenging problem.
It requires the chip to be able to process information in multiple directions at once, with each direction being processed at a different speed.
The researchers developed a way to create a transistor that could do this, but also process data at a lower rate.
“It was very challenging to design a chip that would do that, but the way that we do it is that we use gates, and gate-based transistors are not going to be good at low frequencies,” Chopra says.
“In fact, they don’t work very well at high frequencies, because they’re just not efficient at low frequency.
And it’s not just the data that’s slow.
We can’t have high-frequency data.”
To achieve their theoretical 10-fold performance, the team used a combination of high-quality silicon and low-quality components.
They used a high-density, copper-based die that has a lower electrical conductive index, and used a large number of transistors that are much smaller than a typical chip.
These transistors were fabricated using a method called “gating” — a technique where the number of gates is increased by the number that are on the chip.
This increased number of gate gates allows the processor to process faster information.
In this way, the chip can process data in one direction at once.
The next step was to use a technique called “saturation,” which involves the use of an increase in the temperature of a material that is used as the gate.
This increases the number and size of transposable elements, which can then be added to the chip’s semiconductor.
By adding more transistors on top of each other, the transistor is able to move more information.
“These new gates can go up to a billion transistors,” says Arjun Narayanan, a professor in the MIT Department of Mechanical Engineering and a co-author of the study.
“This is a 10-to-1 increase.
That’s very significant.”
The researchers also used a method known as “saturating” to achieve their 10,600-fold theoretical power.
By increasing the temperature by one-third, they were able to increase the number, and to create more gates and transistors in parallel.
“By increasing the transistors at a higher temperature, we get to a very high efficiency,” Chopras says.