According to a university press release, the world’s first in-memory processor is now available. This large-scale processor will redefine the use of energy to a higher efficiency level when it is processing data. Researchers at LANES or the Laboratory at Nanoscale Electronics and Structures in Switzerland, at the EPFL or Ecole Polytechnique Fédérale de Lausanne have developed the new processor.
The latest information technology systems produce copious amounts of heat. Engineers and scientists are looking for more efficient ways of using energy to lower the production of heat, thereby helping to reduce carbon emissions as the world aims to go greener in the future. In trying to reduce the unwanted heat, they are going to the root of the problem. They want to investigate the von Neumann architecture of a processor.
In a contemporary computing architecture, the information processing center is kept separated from the storage area. Therefore, the system spends much of its energy in shuttling information between the processor and the memory. This made sense in 1945, when John von Neumann first described the architecture. At the time, processing devices and memory storage were intentionally kept separate.
Because of the physical separation, the processor must first retrieve data from the memory before it can perform computations. The action involves movement of electric charges, repeatedly discharging and charging capacitors, including transiting currents. All these leads to energy dissipation in the form of heat.
At EPFL, researchers have developed an in-memory processor, which performs a dual role—that of processing and data storage. Rather than using silicon, the researchers have used another semiconductor—MoS2 or molybdenum disulphide.
According to the researchers, MoS2 can form a stable monolayer, which is only three atoms thick, and can interact only weakly with its surroundings. They created a monolayer consisting of a single transistor, simply by peeling it off using Scotch tape. They could design a 2D version of an extremely compact device using this thin structure.
However, a processor requires many transistors to function properly. The research team at LANE could successfully design a large-scale transistor that consists of 1024 elements. They could make this entire structure within a chip of 1×1 cm dimensions. Within the chip, each component serves as a transistor and a floating gate to store a charge. This controls the conductivity of the transistors.
The crucial achievement of the researchers was the processes the team used for creating the processor. For over a decade, the team has perfected their ability to fabricate entire wafers that had MoS2 in uniform layers. This allowed them to design integrated circuits using industry standard tools on computers. They then translated these designs into physical circuits, leading to mass production of the in-memory processor.
With electronics fabrication in Europe needing a boost for revival, the researchers want to leverage their innovative architecture as a base. Instead of competing in fabrication of silicon wafers, the researchers envisage their research as a ground-breaking effort for using non-von Neumann architecture in future applications. They look forward to using their highly efficient in-memory processor for data-intensive applications, such as those related to Artificial Intelligence.
