A 2D in-memory processor containing more than 1,000 transistors

A 2D in-memory processor containing more than 1,000 transistors

A new chip architecture

EPFL researchers have created a power-efficient in-memory processor using MoS2It collects more than 1000 transistors. This processor, which performs vector matrix multiplication efficiently, represents a shift away from the traditional von Neumann architecture and could boost the European semiconductor industry.

Developed by researchers from EPFL, it is the first large-scale in-memory processor using 2D semiconductor materials, and could significantly reduce the energy footprint of the ICT sector.

When information and communications technology (ICT) processes data, it converts electricity into heat. Already today, CO is the global ICT ecosystem2 The footprint rivals that of aviation. However, it turns out that a significant portion of the energy consumed by computer processors does not go toward performing calculations. Instead, the bulk of the energy used to process data is spent transferring bytes between memory and the processor.

In a research published on November 13 in the journal Nature electronicsResearchers from EPFL’s Faculty of Engineering in the Laboratory of Electronics and Nanostructures (LANES) present a new processor that addresses this inefficiency by integrating data processing and storage in a single device, a so-called memory processor. They broke new ground by creating the first in-memory processor based on a 2D semiconductor material with more than 1,000 transistors, a major milestone on the path to industrial production.

2D semiconductor with 1000 transistors

In a paper published in the journal Nature Electronics, researchers from EPFL’s Faculty of Engineering in the Laboratory of Electronics and Nanostructures (LANES) present a new processor that addresses the inefficiency by integrating data processing and storage into a single device, a so-called in-memory processor. They broke new ground by creating the first in-memory processor based on a 2D semiconductor material with more than 1,000 transistors, a major milestone on the path to industrial production. Credit: 2023 EPFL/Alan Herzog

Von Neumann Legacy

According to Andras Kiss, who led the study, the main reason behind the inefficiency of today’s CPUs is the universally adopted von Neumann architecture. Specifically, the physical separation of components used to perform calculations and store data. Because of this separation, processors need to retrieve data from memory to perform calculations, which involves transferring electrical charges, charging and discharging capacitors, and transmitting currents along lines—all of which dissipate energy.

Until about 20 years ago, this architecture made sense, as different types of hardware were needed to store and process data. But von Neumann architecture is increasingly challenged by more efficient alternatives. “Today, there are ongoing efforts to integrate storage and processing into more comprehensive in-memory processors that contain elements that act as both memory and transistor,” Case explains. His lab has been exploring ways to achieve this goal using molybdenum disulfide (MoS2), a semiconductor material.

A new 2D processor architecture

in Nature electronics Research paper presented by Guilherme Migliato Marega, PhD Assistant at LANES, and his co-authors MoS2An in-memory processor dedicated to one of the basic operations in data processing: matrix-vector multiplication. This process is ubiquitous in digital signal processing and implementation of artificial intelligence models. Improvements in their efficiency could lead to significant energy savings across the entire ICT sector.

Their processor combines 1,024 elements into a chip about one in length. Each element includes 2D MoS2 The transistor, in addition to a floating gate, is used to store charge in its memory, which controls the conductivity of each transistor. Coupling processing and memory in this way radically changes how the processor performs calculations. “By adjusting the conductance of each transistor, we can perform analog vector matrix multiplication in a single step by applying voltages to our processor and measuring the output,” Case explains.

A big step closer to practical applications

Material selection – MoS2 – Played a vital role in the development of the in-memory processor. For one, Moss2 It is a semiconductor – a requirement for the development of transistors. Unlike silicon, which is the semiconductor most commonly used in computer processors today, MoS2 It forms a stable monolayer, only three atoms thick, that interacts only weakly with its surroundings. Its thinness makes it possible to produce very compact devices. Finally, it’s a material that the Case lab knows well. In 2010, they created their first single MoS2 Transistor using a single layer of material peeled off the crystal using scotch tape.

Over the past 13 years, its operations have matured significantly, with Migliato Marega’s contributions playing a key role. “The major advance in moving from one transistor to more than 1,000 has been the quality of the materials we can deposit. After much process optimization, we can now produce entire wafers covered with a homogeneous layer of uniform MoS2. “This allows us to adopt industry standard tools for designing integrated circuits on a computer and translate these designs into actual circuits, opening the door to mass production,” Case says.

Revitalizing the European chip industry

Aside from its purely scientific value, Case sees this result as a testament to the importance of close scientific cooperation between Switzerland and the EU, particularly in the context of the European Chip Act, which aims to strengthen Europe’s competitiveness and flexibility in the field of semiconductor technologies and the semiconductor industry. Applications. “EU funding was crucial for both this project and those that came before it, including the one that funded work on the first MoS2 transistor, showing how important it is for Switzerland,” says Case.

“At the same time, it shows how the work being carried out in Switzerland can benefit the EU as it seeks to revitalize electronics manufacturing. Instead of fighting the same race as any other country, the EU could, for example, focus on developing processing architectures.” Von Neumann changed AI accelerators and other emerging applications.By defining its own race, the continent could get a head start on securing a strong position in the future.

Reference: “A large-scale integrated matrix-vector multiplexing processor based on single-layer molybdenum disulfide memories” by Guilherme Migliato Marega, Hyun-Joo Ji, Chenyu Wang, Gabriele Pasquale, Mukesh Tripathi, Aleksandra Radinovic, and Andras Kiss, 13 November 2023, Nature electronics.
doi: 10.1038/s41928-023-01064-1

You may also like...

Leave a Reply

Your email address will not be published. Required fields are marked *