The latest paper published by Nature shows that researchers at Intel and the University of California at Berkeley are researching super chips, already in "spintronics" Breakthrough in the field of learning.
At present, Moore's Law is “the number of components that can be accommodated in a semiconductor chip, which is doubled in about 18 months”, and the “components” are mainly CMOS. Transistors, the current mainstream view is that after 5nm node, the transistor will approach the physical limit, leading to the end of Moore's Law.
The researchers use "spintronics" technology to shrink the size of today's common chip components to one-fifth and reduce energy consumption by more than 90%. Commercial success is expected to develop a "super chip" to "renew" for Moore's Law.
The transistor technology invented 70 years ago is now widely used in various fields such as mobile phones, electrical appliances, automobiles and supercomputers. The transistors move electrons around the inside of the semiconductor and store them as binary information 0 and 1.
Since the early 1980s, most electronic products have relied on the use of CMOS transistors. However, the principle of CMOS operation involves the switchable semiconductor conductivity controlled by the insulated gate, which is largely constant, even if the transistor can be shrunk to a size of 10 nanometers.
Nature’s research by Intel and the University of California at Berkeley: Exceeding CMOS's scalable logic technology can improve the efficiency and performance of the von Neumann architecture, and Growth in emerging computing such as artificial intelligence.
Specifically, the researchers propose a scalable spintronic logic device "MESO device" that uses spin-orbital transduction and magnetoelectric switches. jobs. The device uses advanced quantum materials, especially related oxides and material topologies.Conduct collective switching and testing.
MESO is based on a multi-iron material consisting of bismuth, iron and oxygen (BiFeO 3 ), both magnetic and ferroelectric. A key advantage of this material is that the two states are interconnected or coupled, so changing one state affects the other. By controlling the change and flip of the electric field, the state of the magnetic field can be changed, which is crucial for the birth of MESO.
Based on magnetoelectric and spin-orbital materials, MESO changed from the binary representation of the original CMOS to the high and low states of the magnetic spin of multiferroic materials. .
Figure 1 : MESO Logical Transduction and Device Operation
Single crystal structure of bismuth iron oxide multiferroic material. Helium atom (blue) is on each side of the cube The oxygen atom (yellow) forms a cubic lattice, and the iron atom (gray) is at the center. The slightly off-center iron interacts with oxygen to form an electric dipole (P), which is coupled with the magnetic spin of the atom (M). The flipping of the magnetic moment also causes the magnetic moment to flip. The common magnetic spin of the atoms in the material encodes the binary information 0 and 1, and implements information storage and logic operations.
< Compared with CMOS technology, MESO has superior conversion energy (10 to 30 times), lower switching voltage (5 times) and enhanced logic density (5 times). In addition, its non-volatile nature Achieving ultra-low standby power consumption is critical for modern computing. This shows that spintronic logic technology can realize the development of multiple generations of computing.
The logic of MESO The operation speed is five times higher than CMOS.The progress of the calculation of the unit area in Moore's Law is continued.
Under "spintronics" technology, MESO is expected in the future replace the widely used CMOS transistors, "super chip" will be born, we can say, MESO is expected to Moore's Law "Continue another wave."
The material for the MESO device was first discovered in 2001 by Ramamoorthy Ramesh, a professor of materials science and engineering and physics at the University of California, Berkeley, and he is also a senior author of the paper.
Ramamoorthy Ramesh believes that the future,There are two major trends in the global computing market that urgently require more energy-efficient computers. One is the Internet of Things and the other is AI.
The Internet of Things means that every building and every car will be fully equipped with microelectronics and everything connected. Although the exact size of this market is being debated, it is agreed that it is growing rapidly.
The artificial intelligence/machine learning is in the early stages, but it will be applied in various technical fields in the future. However, these applications are currently limited by memory limitations and computational efficiency. Therefore, we need a more powerful chip that consumes less energy. Driven by these emerging applications, the microelectronics market is likely to grow exponentially.
Ramamoorthy Ramesh also mentioned that international competition is also the driving force behind the development of next-generation semiconductor technology. At present, China has invested hundreds of billions of dollars in building fabs, which were previously only available to US companies. In the past two years, the fastest computers in the world have been made in China, so this is a strategic issue for the United States.
According to the US Department of Energy, as the computer chip industry expands to trillions of dollars in the next few decades, the percentage of energy consumed by computers will 3% of total energy consumption in the United States has soared to 20%, almost equal to the total energy consumption of today's transportation.
In the paper, the researchers said they have reduced the voltage required for magnetoelectric control switches for multiferro materials from 3 volts to 0.5 volts and predicted the future. It should be able to drop to around 0.1 volts: this is only one-fifth to one-tenth of the currently widely used CMOS tube. Low voltage means low power consumption: the total energy required to represent a binary number using a MESO device is only one tenth to one-thirth of the energy required by CMOS.
However, MESO devices still have a lot to go. Ramamoorthy Ramesh gave a timetable: it will take ten years.
Is the decade too long? In fact, the industry has been very anxious.
The argument that Moore's Law is about to end or even has ended in recent years has become more and more "in-depth", like Nvidia CEO Huang Renxun, who recently said "Moore" in GTC Suzhou not long ago. The law is over."
However, the industry has never really died of prolonging Moore's Law.
IRDS (International Roadmap for Devices and Systems) is an organization established by the IEEE. Since 1965, it has published a technical roadmap for the semiconductor industry every year.Formerly known as the ITRS (International Technology Roadmap for Semiconductors) roadmap, the name was changed to IRDS in 2016 to fully reflect various system-level new technologies.
In 2017, the roadmap released by IRDS triggered an uproar because it predicted that traditional CMOS integrated circuits would be 2024 years old - only six years later - Come to the end.
It can be seen from the IRDS roadmap that starting from 2024, although the semiconductor process will still have 2.5nm and 1.5nm linewidth, pay attention to red. In the frame part, the gate distance and other indicators of these new processes are unchanged, that is to say, the transistor does not shrink, and will not change after the 5 nm node. In other words, traditional CMOS circuits will come to an end in 2024.
However, the CMOS circuit "bumps" in 2024 does not mean that semiconductor technology will stop developing.
The IRDS white paper points out new directions, including the use of new semiconductor materials and manufacturing processes to reduce transistor feature size (also known as "More Moore"), using innovative system integration techniques such as 3D stacking. (More than Moore), and Beyond CMOS - Enhance IC performance with new devices other than CMOS.
Simply, the main idea of Beyond CMOS is to create "new switches" to process information. Features of such devices include, but are not limited to, high functional density, higher performance, lower power consumption, and enough Stable, cost-effective, and capable of large-scale manufacturing.
Beyond CMOS is a hot current academia and industry,At present, there are no fewer than a dozen programs that are vigorously explored, and Intel has spared no effort in this regard.
Intel has multiple paths on Beyond CMOS (below), of which MESO is a recent breakthrough.
Intel believes that MESO devices can reduce voltage requirements by a factor of five compared to existing CMOS solutions, reducing power consumption by 10-30 times in specific situations.
The core point is that MESO is using quantum materials at room temperature, compared to current architectural innovations such as dedicated chips (DSA), from CMOS to If the path of MESO can be realized, it will be a qualitative leap.