background

Transistor is one of the greatest inventions in modern human history. Modern electronic devices such as computers, smart phones and intelligent hardware are inseparable from transistors. After the emergence of integrated circuit technology, a large number of transistors can be packaged in a chip the size of a fingernail. This transistor consists of a source, a drain and a gate between them. The current flows from the source to the drain, and the gate controls the current switch.

The famous Moore's Law points out: "Under the condition of constant price, the number of transistors that can be accommodated in an integrated circuit will double every 18 months, and the performance will also double." As predicted by Moore's Law, the size of transistors is shrinking, and the number of transistors integrated on a single chip is increasing, which makes it possible to perform more and more complex operations.

But in recent years, Moore's law is facing severe challenges. Traditional transistors are mainly made of silicon material. For silicon transistors, 7 nanometers is the physical limit. Once the size of the silicon transistor is lower than this figure, due to the "quantum tunneling effect", electrons will no longer be bound by Ohm's law and cross the barrier that could not have been crossed. This will cause leakage of the integrated circuit and make the transistor unreliable.

In order to solve the above problems and maintain the vitality of Moore's Law, people of insight in industry and science began to actively look for new materials. The goal of these materials is to replace silicon and produce a new generation of transistors with smaller size, better performance and lower power consumption.

For example, the author once introduced that Lawrence Berkeley National Laboratory has developed the world's smallest transistor using carbon nanotubes and molybdenum disulfide, and its transistor manufacturing process is only 1 nm.

For another example, studies by McGill University in Canada and the University of Montreal show that black phosphorus is expected to be a very good candidate material for transistors. In addition, other two-dimensional materials, such as graphene, hexagonal boron nitride, tungsten selenide and so on. , can be used to make new transistors.

The fabrication of carbon nanotube field effect transistor (CNFET) has become one of the main goals of building a new generation of computers. The research shows that the energy efficiency of CNFETs is ten times that of silicon, and the running speed is faster. However, in mass production, these transistors usually have many defects that affect their performance, which is unrealistic.

The microprocessor is based on RISC-V open source chip architecture, which has a set of instructions that the microprocessor can execute. The microprocessor designed by the researchers can accurately execute all instruction sets, and can also execute the classic "Hello, World! hello,world ) "program, printed out:" Hello, world! I am RV 16 x nano, made of carbon nanotubes. Hello, world! I'm RV 16XNano, made of carbon nanotubes. )"。

Max M. Shulaker, assistant professor in the Department of Electrical Engineering and Computing Science of EECS, member of Microsystems Technology Laboratory and co-author of the paper, said: "So far, this is the most advanced chip manufactured by emerging nanotechnology, and it is expected to achieve high-performance and energy-efficient computing. Silicon has limitations. Therefore, if we want to continue to make progress in the field of computing, carbon nanotubes are one of the most promising ways to overcome these limitations. This research paper has completely changed the way we make chips with carbon nanotubes. "

This microprocessor was developed on the basis of an iterative version designed by Shulaker and other researchers six years ago. At that time, the version only had 178 CNFETs, which could only run on single-bit data. Since then, Schuraker and his colleagues at MIT have begun to deal with three unique challenges in the manufacturing process of carbon nanotube microprocessors: material defects, manufacturing defects and functional problems. Gage Hills is responsible for most of the processor design work, while Christian Lau is responsible for most of the manufacturing work.

Schuraker said that the inherent defects of carbon nanotubes have been the "bane" in this field for many years. Ideally, CNFETs need semiconductor characteristics to turn their conductivity on or off, corresponding to whether the bit is 1 or 0, respectively. But inevitably, a small amount of carbon nanotubes will be metallic, which will slow down or prevent the transistor from switching. In order to avoid these failures, advanced circuits will require carbon nanotubes with a purity of 99.999999%, which is almost impossible to produce today.

The researchers put forward a technology called Dream (short for design to resist the elasticity of metal carbon nanotubes). This technique places metal CNFETs in a position that will not interfere with the calculation. In this process, they relaxed the strict purity requirement by four orders of magnitude, namely 1000 times, which means that they only need carbon nanotubes with a purity of 99.99%, and they can prepare them now.

Basically, designing a circuit requires a library of different logic gates connected to transistors, which can be combined to create adders and multipliers, just like splicing letters into words. The researchers found that metal carbon nanotubes have different effects on different combinations of these logic gates. For example, a single metal carbon nanotube in logic gate A can disconnect the connection between logic gate A and logic gate B ... but several metal carbon nanotubes in logic gate B will not affect their connection.

In chip design, there are many ways to realize the code on the circuit. The researchers simulated looking for all different combinations of logic gates, which may be "robust" or "not robust" for any metal carbon nanotubes. Then, they customized a chip design program to automatically find the combination that is least likely to be affected by metal carbon nanotubes. When designing a new chip, the program will only use the "robust" combination and ignore the defective combination.

Schuraker said: "The pun' dream' is very meaningful because it is the solution that everyone dreams of. This method allows us to buy ready-made carbon nanotubes, put them on the chip, and construct our circuits as usual without doing anything special. "

The fabrication of CNFET begins with the deposition of carbon nanotubes on a wafer with a pre-designed transistor structure in solution. However, some carbon nanotubes will inevitably stick together at random to form large bundles, just like spaghetti strung into small balls, forming large particles of pollutants on the chip.

In order to remove this pollutant, the researchers invented the washing technology (removing hatched nanotubes by selective exclusion). The wafer will be pretreated by a reagent that promotes the adhesion of carbon nanotubes. Then, the wafer is coated with a layer of polymer and immersed in a special solvent. In this way, polymers can be washed away, and these polymers can only take away large bundles of carbon nanotubes, while individual carbon nanotubes will still adhere to the wafer. Compared with other similar methods, this technology can reduce the particle density on the chip by about 250 times.

Finally, the researchers solved the common functional problems of CNFET. Binary calculation requires two types of transistors: "N" transistor, where on stands for 1 and off stands for 0; The "P" transistor is the opposite. Traditionally, it is a challenging task to manufacture these two types of transistors with carbon nanotubes, because transistors with different properties are usually produced. In order to solve this problem, the researchers developed a technology called mixed (metal interface engineering and electrostatic doping intersection), which can accurately adjust the function and optimize the transistor.

In this technology, they attach some metals (platinum or titanium) to each transistor to fix the transistor as P or N. Then, they coat CNFET on an oxide compound through atomic layer deposition, thus adjusting the characteristics of the transistor to meet the needs of specific applications. For example, servers usually need transistors that run fast but consume a lot of power. On the other hand, wearable devices and medical implants may require slower and lower power transistors.

future

Their main goal is to push the chip into the real world. In order to achieve this goal, researchers have now begun to apply their manufacturing technology to a silicon wafer foundry through a project of the Advanced Research Projects Agency of the US Department of Defense, which supports this research. Although no one can be sure when chips made entirely of carbon nanotubes will be available. But Schuraker said: "It may be realized within five years. We think this is not a question of whether it can be realized, but only a question of when it will be realized.

key word

reference data

1Gage Hills, etc. Modern microprocessors constructed by complementary carbon nanotube transistors, Nature (20 19). DOI: 10. 1038/s 4 1586-0 19- 1493-8

2 http://news.mit.edu/2019/carbon nanotubes-microprocessor -0828.