Get best of the week

How FinFET's father helped save Moore's law

Chenming Hu, 2020 IEEE Medal of Honor, brings transistors to the third dimension


It was 1995. Advances in chip technology have not lagged behind Moore’s law — the observation that the number of transistors in a chip doubles approximately every two years — mainly due to a decrease in the size of transistors.

However, the horizon no longer seemed so limitless. For the first time, rumors have spread throughout the semiconductor industry that predict the demise of Moore's law. The gold money will come to an end, predictors broadcast, when the size of the critical characteristics of the transistor, which then had a size of the order of 350 nm, drops to 100 nm. Even the US government was worried - so much so that the DARPA agency sounded the alarm and launched a program to search for new chip technologies that could continue progress.

Chenmin Hu, then a professor of electrical and computer science at the University of California at Berkeley, was happy to accept this challenge. He immediately came up with a solution to the problem - in fact, even two - and, sitting on the plane, sketched them in draft. One of these ideas was to raise the current channel so that it rises above the surface of the chip. It turned into FinFET technology, for which this year Hu was awarded the IEEE Medal of Honor "for an outstanding career in the development and practical application of semiconductor models, in particular three-dimensional structures, which contributed to the preservation of Moore's law for many decades."

Naturally, the FinFET story did not begin with the fact that Hu began to draw something with a pencil on paper on a folding table in an airplane.

It began in Taiwan, where Hu, a curious child, conducted home experiments with sea water, and dismantled (and then collected) alarms. After graduating from school, he was still interested in science, mainly chemistry. But instead of studying to be a chemist, he entered the Faculty of Electrical Engineering at National Taiwan University , without even knowing exactly what electrical engineering does. It was just a challenge for him - for this training program, the highest marks were required.

In his final year of study, Hu discovered an industry that would be shocked by his actions - all thanks to Frank Fang, who was invited to lecture from the United States.

“It was 1968,” Hu recalls, “and Fang told us that future televisions will be based on semiconductors, and that television sets will turn into something like photos that can be hung on walls.”

In the era of bulky TVs with picture tubes, this attracted Hu's attention. He decided that studying semiconductors would suit him, and applied for training in the United States. In 1969, he ended up in Berkeley, where he joined a research team working on metal oxide semiconductor transistors ( MOS structures ).

Soon his career changed direction, because, as he recalls, it seemed to him too easy. He switched to the study of optical circuits, defended his doctorate in integrated optics, and transferred to MIT to continue work in this area.

And then, in 1973, an oil embargo was introduced. “It seemed to me that I should do something important,” he said, “useful, and not just write some work.”

Therefore, he switched to the development of inexpensive solar panels for ground-based applications - in those days, solar cells were used only on satellites. In 1976, he returned to Berkeley, already a professor, planning to conduct research in the field of energy, including hybrid cars - and they led him back to semiconductors. “Electric cars,” Hu explains, “require high voltage and high current semiconductor devices.”

In the early 1980s, a return to semiconductor research was a coup. The government stopped funding energy research, but several companies located in the San Francisco Bay supported semiconductor research and switching to corporate finance “wasn’t very difficult to implement,” says Hu. He began spending more time in Silicon Valley, near Berkeley, at the invitation of companies taught short courses on semiconductor devices. In 1982, he spent his holidays in the heart of Silicon Valley with the National Semiconductor in Santa Clara.

“My involvement in this industry has long affected my life,” says Hu. - In science, we learn important things from each other, and I was interested in something when I read someone else's work and thought that I could do better. And then, when I got acquainted with the industry, I realized that it was there that interesting tasks lurked. ” And this revelation has led Hu to become more active in exploring the three-dimensional structures of transistors.


FinFET features: each transistor has a source, a drain, a conductive channel connecting them, and a gate that controls the current in the channel. In FinFET, the channel is raised above the surface of the chip - like a shark fin. fin - fin] - which allows the shutter to wrap it on three sides, as a result of which it can better control the current.

A field-effect transistor has four main components - a source, a drain, a conductive channel connecting them, and a gate that controls the current in the channel. And the less these components were made, the more often people noticed changes in the behavior of transistors after a long time. These changes did not manifest themselves in short-term trials, and companies had difficulty predicting whether they would manifest or not.

In 1983, Hu read a paper published by IBM researchers that described a similar change. Thanks to his experience at National Semiconductor, he realized what difficulties the industry might face due to this lack of long-term reliability. If he didn’t work “at the forefront,” he says, “I would not understand the importance of this problem, and would not want to spend almost 10 years on solving it.”

Hu decided to accept the challenge, and with a group of students he developed the theory of hot carrier injection to predict the reliability of MOS. This numerical model describes the degradation of a device during electron migration. Then he turned to another problem of reliability: how oxides degrade over time, which became important as manufacturers gradually made oxide layers of semiconductors more and more thin.

Hu says that these studies required him to carefully understand the processes taking place inside transistors. Subsequently, this work gave rise to the Berkeley Reliability Tool (BERT) and BSIM transistor model sets. BSIM has become the industry standard and is still in use today. Hu is still in charge of regularly updating his models.

Hu continued to work with students, studying the basic characteristics of transistors - how they work, how they fail, how they change over time - until the mid-1990s. Meanwhile, commercial chips developed in accordance with Moore's law. However, by the mid-90s, when the average characteristic size reached 350 nm, prospects for further compressing the size of transistors began to cause concern among manufacturers.

“The end of Moore’s law is already visible,” recalls Lewis Terman, who was working at IBM Research at the time.

The main problem was nutrition. The smaller the characteristic dimensions, the more problems caused the current flowing away while the transistor is closed. And these leaks became so significant that they increased - or even accounted for most - of the chip’s power consumption.

“Work began to appear with predictions that the Moore’s law for CMOS would end when the 100 nm threshold was crossed, because at some point it would have to dissipate more energy per square centimeter than in the nozzle of a rocket,” recalls Hu. “The industry has declared the battle lost.”


Chenming Hu began teaching at the University of California at Berkeley in 1976.


Among the first topics of his research were hybrid cars, in particular, a gasoline-electric car, which he presented at a meeting of board members of the University of California in 1980.


In his laboratory in 1997, Hu was actively involved in the development of FinFET with money from DARPA.

Not wanting to abandon Moore's law, DARPAThe US Department of Defense Advanced Research Projects Office was looking for studies that promised to overcome this barrier in order to finance them. In mid-1995, it launched a project called the “25 nm transition.”

“I liked the idea with dimensions of 25 nm - it went far enough beyond what was considered possible in the industry,” Hu said.

Hu considered the fundamental problem to be extremely clear - you need to make the channel so thin that the electrons could not slip past the gate. At that time, among the solutions to this problem was the proposal to make the oxide layer in the gate thinner. Thanks to this, the control of the channel became better, and the leakage current decreased. However, Hu's work showed that such an approach was too close to the dangerous limit: if the oxide layer was made too thin, the electrons would be able to jump through it onto a silicon substrate, which would give rise to another source of leakage.

Two other options immediately crossed my mind. One is to complicate the charge around the gate by adding an insulation layer in silicon under the transistor. Such a circuit was called “fully depleted silicon on a substrate”, or FDSOI. Another was to increase the shutter's ability to control charge by lifting a thin channel above the substrate in the manner of a shark fin - so that the shutter could be wrapped around the channel from three sides, and not just held from above. This structure was called FinFET, and it had another advantage - the use of the third dimension reduced the load on the two-dimensional plane, and paved the way for the creation of three-dimensional transistors.

However, there was not much time left to send the application to DARPA. Hu learned about the funding proposal from one of his colleagues, Jeffrey Bokor, who found out about it while windsurfing with the program director from DARPA. Therefore, Hu quickly met with Bokor and another of their colleagues, Tsu Jae King, and agreed that the team would draw a proposal for the week. After a couple of days, while on a plane flying to Japan, he sketched two versions of the scheme, and upon arrival at his hotel sent the drawings and technical description by fax back to Berkeley. The team sent his proposal, and later DARPA allocated a four-year research grant to the team.

Until then, ideas similar to FinFET had already appeared in scientific papers. However, Hu and his team actually created devices suitable for industrial production, and showed how their circuitry will make transistors with a characteristic size of 25 nm or less. “Other scientists who read these works did not consider such an approach a solution to the problem, since such transistors would be difficult to manufacture, and it is unclear whether they will work or not. And even the authors of the works themselves did not develop this idea further, says Hu. - I think that the difference was that we looked at this problem and decided that we want to work with it not because we wanted to write another job or get another grant, but because we wanted to help the industry. We thought we needed to extend the Moore’s law. ”

“We, as technologists,” Hu continues, “were responsible for ensuring that he did not stop. Because as soon as he stops, we will immediately lose hope of expanding our capabilities to solve the most complex problems of mankind. ”


, 12 1947 .
180.
, .
, 1968.
, 1973 – .
: .
: - MIT.
: .
100 .
: .
ACM Research, Ambarella, Inphi.
: [ TSMC, ].
: Psychology Today.
: .
: (1993)
: , .
. Tesla S.
: IEEE 2020 « , , , », .


He and the team “were well suited to developing FinFET because he taught his students how to think about devices,” says Elise Rosenbaum, his former student, and now a professor at the University of Illinois at Urbana-Champaign. “He emphasizes the importance of the overall picture, a quantitative understanding of the situation. Studying a semiconductor device, some people concentrate on creating a model and subsequent numerical solution of all points of a three-dimensional grid. He taught us to take a step back, try to imagine the distribution of the electric field in the device, the location of potential barriers, and how the electric current changes when we change the size of a certain part. "

Hu believed in the importance of visualizing the behavior of semiconductor devices so much that once, according to Rosenbaum, trying to teach students this process, "he built for us a model of the behavior of a MOS transistor from plasticine taken from his children."

“It all looked like an invention that suddenly appeared from scratch,” said Fari Assaderagi, a former student and now vice president of innovation and technology at NXP Semiconductors . - However, his team worked on the fundamental concepts of an ideal device, starting with the very foundations of physics. And the idea of ​​creating such a structure comes from there. ”

By the year 2000, at the end of four years of financial support, Hu and his team created working devices and published their research, which attracted the instant interest of many industry representatives. However, it took another ten years for FinFET chips to begin to roll off the assembly lines, and the Intel chip was the first in 2011. Why did it take so long?

“The situation has not yet broken,” Hu explains, referring to the industry’s ability to produce more and more compact designs. “People thought it would break, but you can't repair something that hasn't broken yet.”

It turned out that the DARPA managers predicted the future - they called the financing project “the transition to 25 nm,” and when FinFET appeared, the semiconductor industry had already switched to technological processes below 25 nm.

Meanwhile, FDSOI has also evolved and is even used today in the industry. In particular, it is used in optical and radio devices, and FinFET dominates the processor industry. Hu says he never said that one approach is better than the other.

In the days of FinFET's inception, Hu took a three-year vacation from Berkeley to work as CTO at semiconductor maker TSMC in Taiwan. He considered it an opportunity to repay a debt to the country where he received his initial education. He returned to Berkeley in 2004, continued to teach, study energy-efficient semiconductor devices, and support BSIM. In 2009, Hu completed regular teaching, but still works with graduate students as an honorary professor.

After Hu returned to Berkeley, FinFET technology captured the industry. And Moore's law did not end at 25 nm, although his death is still regularly predicted.

“Progress will gradually slow down, but we won’t have a replacement for MOS for another hundred years,” says Hu. However, he does not lose hope. “There are ways to improve circuit density, power consumption, and speed, and we can expect the semiconductor industry to continue to provide people with increasingly useful, convenient, and portable devices. We need to approach the issue more creatively and with more confidence. ”

More articles:


All Articles