Chapter 22 Rapid progress
Bavendi is an expert in the field of perovskites, but he won the Nobel Prize in 23 not because of his achievements in perovskites, but because of his innovation in the chemical preparation methods of quantum dots.

Quantum dots are tiny semiconductor particles with unique optical and electronic properties that can be used in LEDs, infrared detection, solar cells and other fields.

When Bawendi heard that the other party chose perovskite, he smiled and said, "Wright, I'm not sure how good you are at doing experiments, because your paper didn't show your talent in this area.

If you want to publish a paper on perovskite, the experimental requirements are very high. You need to be able to produce results that others cannot produce, or observe phenomena that others cannot observe, and be able to theorize them. "

Bawendi continued: "Of course, if you want to publish papers in top journals, I know this is very important for Chinese students, because you need these supporting materials to prove your ability so that you can get better treatment when you return to China.

From this perspective, perovskite is indeed a very good direction. Every year, there are no less than ten papers on perovskite in Nature alone. Together with Science, this is indeed a direction that is easy to produce results.

I'm just not sure how talented you are at doing experiments."

Chen Yuanguang was full of confidence. This was nothing compared to the topic Pangu had given him: "I think I can do it."

Bawendi stared into his eyes for a while, then nodded: "Ok, as long as you think it's okay, then young man, let's get to work!"

After Bavendi sent him the list of papers and related textbooks, he let him study on his own. This is the style of MIT. The tutor assumes that you are a genius.

"Has Chen been to the lab?" Bawendi asked.

An Indian student said: "I haven't seen him."

"Okay." Bawendi was a little confused.

When Chen Yuanguang sent him another email, hoping that he would recommend some new papers, Bawendi replied with an email, asking, “Do you have any ideas for the paper?”

“I have a rough idea. I plan to study how to use machine learning methods to predict the high-throughput effects of antisolvents on perovskite stability,” Chen Yuanguang replied.

After reading it, Bawendi realized that Chen Yuanguang had returned to the path he was familiar with, that of computational chemistry.

Machine learning is great. In 2016, AlphaGo emerged and defeated Go player Lee Sedol, which was hailed as the first year of machine learning.

Two years have passed now, and everyone feels that artificial intelligence should be combined with their major, just like Internet+ was everywhere twenty years ago.

Now it is artificial intelligence+, but artificial intelligence talents are hard to find. The prices offered by big companies such as Google, Amazon, and FB to talents in the field of artificial intelligence are astronomical, and the prices offered by some startup companies are even higher than each other.

It is very difficult to recruit someone who understands both artificial intelligence and chemistry, let alone someone who can combine the two.

Even Bawendi couldn't find talent in this area. The appearance of Chen Yuanguang gave Bawendi some ideas, but he didn't expect that the other party would adapt so quickly and found the point of integration in just one month.

"Wright, come to my office tomorrow and we'll talk about the specific research direction." After seeing Chen Yuanguang's reply, Bawendi stopped sending an email and called directly.

The next day, "Wright, tell me roughly what you think." Bawendi handed the coffee to Chen Yuanguang.

"I think we can combine automated characterization, chemical robotic synthesis technology and machine learning to explore how the choice of antisolvent affects the intrinsic stability of perovskites. For example, we choose different ends to make combinations, such as MAPbI3, CsPbI3 and CsPbBr3, etc., to synthesize some combinatorial libraries, each of which will have its own unique combination.

I expect that we will be able to synthesize more than a thousand components in total, and then make each library twice using two different antisolvents: toluene and chloroform.

After synthesis, photoluminescence spectrum analysis is automatically performed every 5 minutes for a period of time. The appropriate length of time needs to be determined by experiments.

Finally, non-negative matrix factorization is used to map the time- and composition-dependent optoelectronic properties.

By using this workflow for each library, we can find out how the choice of antisolvent affects the intrinsic stability of the perovskite.

In fact, it may be a dynamic process." Chen Yuanguang briefly talked about his ideas.

This idea is very similar to the previous directed evolution of proteins, but the difference is that experiments are done first, a fixed process is set up through chemical robots, and then machine learning is used for analysis.

Bawendi nodded repeatedly, thinking that it was indeed a good idea. No wonder Levitt was resentful when he called him: "Great idea, Wright, I support you.

It depends on the final result. If the result is good, it can be published in Nature or Science.

If the results are not satisfactory, they can still be published in JACS or Nature’s sub-journals. For scholars in the field of chemistry who are behind the times, they will still give face to the most popular artificial intelligence algorithms nowadays.”

Chen Yuanguang thought that it was mainly because the Internet industry paid too well. Those who could combine perovskites and machine learning either had not yet grown up (after all, artificial intelligence has only been popular for two years), or they would not have come to study for a doctorate at all and had already gone to the industry to make a lot of money.

Just because this huge sum of money is nothing to him, he has the opportunity to study here.

"Professor, our laboratory does not have a graphics card. I need to buy the latest graphics card from NVIDIA to do calculations," Chen Yuanguang mentioned.

Bawendi said helplessly: "It seems that I am also out of date. Buy it. Please write me an application form. After I sign it, you can give it to Susan and ask her to go to purchase it."

"It's okay. The result is not good, but not bad either. It may be a bit difficult to publish in Nature, but it is more than enough to publish in JACS." Bawendi stared at the result for a long time and then said.

Chen Yuanguang said: "JACS is good enough. While I was working on this project, I also started another project. This project can definitely be published in Nature."

Bawendi was shocked: "Tell me about it."

Chen Yuanguang: “I built a cross-attribute deep learning framework on GitHub, which can be mainly used for predictive analysis of small material data.

We mainly used some ready-made data from the college to train this model. We first built a large data set, and then built small data sets with different attributes and their models based on this large data set.

Through the framework extracted from these models, we can directly input physical properties as its calculation and experimental data set, and finally obtain its other characteristic results. "

To put it simply, we dug a little deeper into the topic of perovskite prediction and came up with a more general result.

(End of this chapter)