Reborn in 2010, the AI ​​in my mind is too powerful.

Chen Yiran acted immediately, after all, he was now a wealthy man with someone to take care of him, and opening an account was just a matter of making a phone call.

While tossing the ball to Pidan, he took out his phone and called his personal secretary at Datong.

"Hello Mr. Chen, this is Linda. It's a pleasure to assist you. How can I help you?"

Linda is one of Richard's secretaries, specifically responsible for receiving Chen Yiran. She usually handles simple matters, and if she can't handle something, she'll ask Richard for resources.

"Linda, I need to open a stock trading account."

"Okay, Mr. Chen, Stock Brokerage Account Opening. I can process it for you right away. Your investment account will be available on your mobile app in five minutes, and you can freely transfer money into this account."

Do you need our assistance with leveraged financing or other financial derivatives trading? Our investment advisors are highly professional, and you can contact us with any needs you may have.

"No need, no need, I can handle it myself, I can buy it myself."

Chen Yiran waited a while, played tug-of-war with Pidan for a bit, and felt that the time was about right. He opened the mobile app, completed a few verifications, and then the stock market guru Chen Yiran was ready to start investing.

Baidu, $15 million. Amazon, $15 million. Apple, $15 million. Netflix, $15 million. Sina, $15 million. Salesforce, $15 million.

Of course, he couldn't place the order all at once. Although Chen Yiran wasn't an economics major, he was still a college graduate and had studied Economics 101.

Under Jasmine's instructions, Chen Yiran bought the stocks little by little, in batches over the course of the day, avoiding excessive slippage caused by directly raising the stock price. In the end, he managed to secure all the stocks.

In reality, if Chen Yiran were to be pragmatic, he should have gone all in on Sina. If he had bought Sina at that time and sold it in April 2010, he could have directly made a 4-fold profit.

However, since the data he saw in his previous life was not complete, going all in would still be risky, so he simply diversified his investments into these six stocks that he had seen in his previous life.

According to Jasmine's analysis of the icons in her memory, even the least expensive apple has risen by at least 3% this year, and the others have also seen significant increases. If calculated over a year, one could easily double their money back.

If he traded solely based on the highs on the charts, he should have earned $3 million in 2014. Moreover, he possessed a future chart; as long as he didn't buy too much each time and didn't become the butterfly that caused the tsunami, he could profit from these stocks until 2022.

But Chen Yiran still regretted not paying attention to the stock market and various futures markets regularly in her previous life.

All you do is watch games all day—NFL, NBA, F1, Premier League, Champions League, World Cup, PUBG, Valorant, CSL, OL, Dota 2—what's the point? Those stories about winning billions in a World Cup are just for fools; casinos have limits. Not only are there limits on betting, but there are also limits on payouts. Even the biggest casinos will only pay you a maximum of two or three million dollars per game.

Chen Yiran was indignant for a few days, then put the matter out of his mind because his company was finally about to open.

On Children's Day, June 1st, FlowEDA finally officially started work.

Chen Yiran was not a pure atheist in her previous life, and having encountered such a bizarre event as time travel, she decided to believe it was true rather than risk it.

He led a group of employees to first pay homage to Guan Yu, then to Bi Gan, Fan Li, and Zhao Gongming, who, according to Chinese folklore, are all civil and military gods of wealth.

The gods of fortune, prosperity, and longevity, as well as the kitchen god, were not neglected and all received gifts.

Then we prayed to Bodhisattva Maitreya, Jesus, God, and Allah. To accommodate the Indian employees, we also offered prayers to the three main deities.

The key is that he's familiar with all sorts of powerful figures, and wherever he goes, he has someone protecting him.

Although the employees were somewhat speechless about their boss's superstition, they still cooperated and went through the procedures.

After that, the others who hadn't signed contracts all did. Then they found their workstations, all furnished with Aeron chairs, each with two SureFire flashlights and two bottles of SABRE OC-17 spray on it.

Coupled with the company's somewhat exaggerated security doors and the reinforcing ribs on the walls, the company's corporate culture was established from day one.

Technique doesn't have to be the strongest, but martial virtue must be abundant.

As per Chen Yiran's request, the company did not set up a meeting room, as there were only eleven people in total, and a meeting room was really unnecessary.

The four separate offices were given to Gao Dewei, Raj, and Zhou Tianyou, with the last one prepared for the best-performing C++ system engineer. Chen Yiran naturally decided to work from home, only visiting occasionally.

On the first day of business, some mobilization was necessary, so Chen Yiran called everyone to the center of the open workstation area.

Chen Yiran was still a little excited, wondering if these people would stay with him in the future. FlowEDA was just a small stage in his journey. The future head office certainly wouldn't be called that. But he wondered if these eleven people could be called the "Jasmine Eleven Arhats."

Chen Yiran tightened his grip on the company's mythical beast, Pidan, to make him behave, and then began to give instructions.

"Today is my first day at the company. Let me start by saying a few things."

He picked up a marker and wrote three words on the whiteboard: Placement, RL, EDA.

"Our company's first product is a chip placement engine driven by reinforcement learning. Placement is the most critical and painful step in the chip design process. Millions of standard cells and billions of interconnects need to be arranged optimally on a silicon wafer of just a few square centimeters. Traditional methods, such as simulated annealing, force-directed methods, and analytical methods, have been used for more than 20 years, but progress has become increasingly slow."

He tapped the letter RL on the whiteboard.

"Reinforcement learning was a small project I did before. The core of it was a computing framework called DeepFlow. I've used this framework for some things, such as home security. I developed the incremental computing engine, the dirty label propagation algorithm, and the topology-driven dependency management of this framework from scratch."

What are the core advantages of DeepFlow? It's not a typical deep learning framework. Ordinary DL frameworks run the entire computation graph in each iteration, regardless of changes. DeepFlow is different; it remembers the computation results of the previous iteration and only recalculates the truly changed parts. How does this relate to EDA placement?

Gao Dewei continued, "Each perturbation of the placement only changes the position of a few local components; the changes in global line length and timing do not need to be recalculated from scratch. This is precisely what DeepFlow excels at: incremental updates."

"right."

Chen Yiran drew a simple diagram on the whiteboard: "Traditional placement engines re-evaluate the global line length after each component movement, resulting in O(n) complexity. However, using DeepFlow for incremental updates only recalculates the affected local areas, achieving O(log n). This means we can run more iterations with the same computing power, which means AI learns faster."

Gao Dewei added, "Traditional simulated annealing can run tens of thousands of perturbations overnight. With the DeepFlow architecture, we expect to run it millions of times in the same amount of time. Reinforcement learning can only converge under this scale of exploration."

There was a few seconds of silence in the conference room. Then Lin Zhiwei spoke first.

"I work on ICs at Synopsys, and the biggest pain point in QoR optimization and placement for the compiler is the computational cost. Running a full benchmark takes dozens of hours, and changing a single line of code takes another dozens of hours. I can't even imagine if the iteration speed could be dozens of times faster."

"You can think about it now."

As Chen Yiran spoke, he turned around and wrote the division of labor on the whiteboard.

Gao Dewei: Overall Architecture + Placement Algorithm Direction

Lin Zhiwei: Placement Engine Core + Benchmark System

Amit: VLSI Design Constraints + Standard Cell Library Adaptation

Zhou Tianyou: GPU Training Cluster + Distributed Training Architecture

Raj: Importing Experience from VLSI Verification + GPU Architecture Verification

Zhang Minghui/ Liu Yi/ Sang Jie/ Wang Xue: C++ Infrastructure + Toolchain

"Catherine handles administration, finance, and payroll. You can talk to her about any problems you have; if she can't solve them, then talk to me."

Catherine stood up in the corner and nodded to everyone. She was wearing a white shirt today and looked quite stylish.

Chen Yiran put down the marker: "Any other questions?"

Raj raised his hand: "What was our first benchmark?"

Gao Dewei replied, "Cadence's ISPD 2006 benchmark suite. It's an industry-standard placement test suite, ranging from hundreds of thousands to millions of cells. We first get the end-to-end process working before scaling up."

"When will the GPU cluster arrive?" Zhou Tianyou asked.

"The three R710s are on their way and will arrive this afternoon."

Wang Xue raised her hand, a little embarrassed: "What do I need to learn now? I don't think I can start right away."

"The basic concepts of EDA are: what is a standard cell, what is a netlist, and what is a line length model. Professor Gao will provide you with materials. You will have two weeks to get started with the toolchain, and in the third week, you will start writing low-level optimizations."

Several C++ engineers nodded, looking a little nervous.

After the meeting, everyone went to their respective workstations and started working.

Since it was the first day, Chen Yiran didn't skip out immediately, but instead scrolled through her phone in Gao Dewei's office.

In the afternoon, the three R710 units arrived. Zhou Tianyou, along with Chen Yiran and Gao Dewei, went to the server room to connect them. The three of them spent half an hour unpacking them. The items weren't particularly large, but the packaging was outrageous; the room was filled with foam and cardboard boxes.

Chen Yiran called over a few more men to help, and they threw the pile of garbage into the outside trash can.

Then the aroma of coffee wafted out.

Catherine has a good eye and made coffee for everyone. It tastes pretty good and is worth the price of this expensive coffee machine.

After finishing his coffee, Chen Yiran looked at everyone like a supervisor and then slipped away.

He has something important to do today. An old friend flew to San Francisco last night and is coming to Silicon Valley to see him today.