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

This was the first time Chen Yiran had ever seen Jasmine so highly praise a technician, especially one who wasn't particularly well-known.

He envisions creating the Epiphany architecture, which uses a large number of minimalist RISC cores arranged in parallel arrays, a completely different approach from Nvidia. By using a large number of extremely simple cores and efficiently interconnecting them, he aims to achieve terrifying parallel computing power, completely bypassing NVIDIA cards and CUDA, and allowing direct programming in standard C.

While this architecture is complex, it is very suitable for China's specific circumstances. It can effectively bridge the chip gap when breakthroughs in manufacturing precision are not possible in the short term. Its various technological approaches and directions are highly aligned with the development trend of China's chip industry.

Chen Yiran looked up at Olofsson, who was still engrossed in drawing across from him. This Swede had been talking about technology for over ten minutes straight without even drinking water, from Reduced Instruction Set Computing to on-chip network arbitration algorithms, his face beaming and his hands gesturing wildly, which perfectly matched his stereotype of technical personnel.

Chen Yiran interrupted Olofsson, asking, "What's your company's current structure, and how many rounds of financing have you completed?"

Olofsson visibly wilted, paused for a few seconds, and said, "We're still in talks. Our company is still quite well-regarded."

He seems perfectly normal when it comes to technology, but utterly useless when it comes to business—that's a very accurate stereotype.

When technical skills and social skills are inversely proportional to this extent, it can no longer be called being unbalanced; it means that all the talent points are put into the same skill tree, and the rest of the attributes are blank.

How many people are there in your company now?

"Twelve. Two do the architecture, three do RTL, two do verification, and the rest work on the software toolchain. We don't have anyone specifically doing physical design; placement and routing are all handled by open-source tools."

"How are the results after running the open-source code?"

"65 nanometers is barely enough to run. It wastes 15 percent more area than normal."

Fifteen percent means burning tens of cents more per chip, which adds up to a lot of money when done in large quantities. But for a team that can't even afford placement tools, discussing wasted space is already a Versailles-like topic.

Let's get him, let's first win him over a little, gain his favor. We'll deal with him when he's down on his luck. I haven't heard of him in any of my past life's information; he certainly wasn't particularly successful, probably strangled by Nvidia.

Chen Yiran felt it was normal for him to be eliminated. Olofsson was clearly a pure technical talent, and under Nvidia's well-established commercial pressure, he could probably only achieve some academic success.

"How do you plan to resolve the placement issue?"

"I was thinking that if we could raise a Series B round and have enough money, I'd buy a Synaptics chip to try it out if needed. If it doesn't work, that's fine too; it's open source, so it'll be usable."

Chen Yiran almost couldn't help but laugh; Olofsson's expression was truly hilarious.

The pricing logic for EDA software is incredibly complex, and it's somewhat opportunistic. Firstly, because chip manufacturing processes change every year, EDA tools need to be upgraded annually to adapt to the new process nodes.

Therefore, it is almost never a one-time buyout, but rather a contract signed annually with an annual licensing fee.

The specific charges are based on the number of seats, with each seat representing an authorization for one engineer.

Synaptics can offer discounts of up to 30% or even lower to major clients like Qualcomm and Intel, but a small company like Adapteva with only twelve employees isn't even considered a client in their eyes. Their sales team might not even bother typing a single word when they receive your inquiry email; they'll just send you a standard template as an attachment, without offering any special treatment.

The publicly listed price for a full-featured placement kit is 150,000 per seat. This kit includes placement, timing analysis, and power optimization. There is also a minimum order requirement, usually five seats.

A single-target tool like Grey, which only performs placement, would cost around 70,000 to 80,000 yuan a year.

Olofsson, as a veteran in the industry, certainly knows the price and that they absolutely cannot afford it; he's just being stubborn by talking about it now.

Chen Yiran first discussed it with Jasmine: "How can we just give him all this for free to win favor? Are you sure his skills are really that good?" I couldn't tell; it seemed like combining a bunch of smaller things into one big one wasn't a very difficult idea.

It's not a simple matter of approach; it's about the actual design concept. Look at the structure he drew on paper—it's very logical. I think we can offer him a discount, but not for free. We still need to make him spend money, and try to get him to use up the company's funds as soon as possible.

His resume strongly suggests that he likely funded most of his companies himself. We've already reached an agreement with Nvidia to prioritize support for CUDA on NVIDIA cards, which will make it extremely difficult for other types of companies to raise funds. Then, hehehehe.

Chen Yiran had a plan in mind. After thinking for a moment, he gave a price that Olofson would probably pay, but it would still be painful for him.

"I'm still very optimistic about your approach. Here's what we'll do: Grey is offering you a special startup rate. $50,000 a year, five slots. No limit on design projects or tape-out times. I'll handle technical issues directly with you, but the response time might be a bit slow, since we're losing money at this price."

Olofson paused for a moment, then quickly took out a somewhat unfashionable business card and exchanged contact information with Chen Yiran.

Chen Yiran and Olofsson chatted for a long time. Chen Yiran mostly followed Olofsson's lead in discussing technical matters, and the two had a very pleasant conversation.

After securing the deal with this small company in Olofsund, although the foot traffic remained high, few people showed any real interest.

This is because Grey was released too hastily, and PDK certification has not yet been completed.

PDK refers to a kit provided by wafer foundries (SMIC, TSMC, Samsung) to EDA manufacturers and design companies, which contains all the physical parameters of the foundry.

Design rules are the basic data, such as minimum spacing between metal wires, through-hole size, and stacking rules.

Device models are how transistors behave at different voltages and temperatures.

The standard cell library is the physical layout of basic logic gates (AND/OR/NOT/flip-flops).

DRC rules and LVS rules are strategies for automatically checking whether a design violates manufacturing rules and whether the circuit diagram and layout are consistent.

These all require EDA vendors to instill them in the software to ensure there are no errors, so that the design company can successfully fabricate the wafer at the foundry.

While Grey's current data is impressive, it hasn't completed any PDK certifications yet. People are just waiting to see what happens, because if it doesn't complete any PDK certifications later, then using it will be a waste of time for enterprises.

As a new company, obtaining certifications from large wafer foundries is extremely difficult.

Take TSMC for example; they are extremely conservative with new EDA vendors. Because once they certify your tools, they also have to take responsibility if problems arise. Therefore, the certification cycle is often ridiculously long, ranging from six months to a year, which can be incredibly frustrating.

Samsung will be a little better, but not by much.

Among the companies that have been confirmed so far, HiSilicon is willing to take the risk because it knows SMIC will cooperate. SiTime is targeting mixed-signal dedicated line wafer fabs, which typically have low barriers to entry.

Olofson Adapteva was simple; he was so poor he was practically selling his pants, so he'd use anything he could get his hands on.

Moreover, many other companies, although they expressed their interest, all said they would contact us again after we obtained more PDK certifications.

On the third day of the exhibition, Chen Yiran handed over the entire booth to Gao Dewei and started to explore the exhibition himself.

First, head straight to the South Building, which is where the manufacturing site is located.

The first thing you see is the Applied Materials exhibition area, which occupies almost half of the venue. The vacuum chamber for chemical vapor deposition is huge, and the wafer transfer robotic arm is also very cool.

These can all be considered high-end excavators, the kind that a man could stare at for ages.

Later, at Lam Research’s booth, Chen Yiran saw the etching machine for the first time.

This is a company that does plasma etching. They put a real machine on their booth, and you could see the demonstration running inside through the viewing window.

The 40-nanometer silicon wafer etching process involves cutting down hundreds of atomic layers per second. The presenter mentioned that this machine has already been sold to TSMC and will be packed up and shipped to Hsinchu immediately after the exhibition.

KLA's testing equipment achieves single-molecule level defect scanning, with nanoscale defect images scrolling on the screen in real time. Each bright spot represents an impurity atom that has landed where it shouldn't.

Chen Yiran stood in front of Applied Materials for a full forty minutes, listening to the entire process of the 40nm process, from polysilicon deposition to metal layer etching. Seeing how attentive he was, the engineer from Lam Research gave him an entire technical white paper, over a hundred pages long, printed on glossy coated paper. It was almost as heavy as a university physics textbook.

SMIC is also working on 40nm, but Chen Yiran only truly understood SMIC's situation after taking a tour of the South Hall.

The gap between TSMC and Samsung is indeed enormous. Even though TSMC is currently mired in the 40nm yield crisis, its technological reserves still far surpass those of SMIC. Samsung has been incredibly aggressive in recent years; aside from its chip manufacturing, it has already entered the 32nm era for flash memory.

If we broaden our perspective to the entire chip industry, the reality becomes even more disheartening.

Synaptics, Cadence, and Mindray, three American EDA companies, account for more than 80% of the global market share.

There's only one domestic company, Huada Jiutian, which was established just over a year ago, and its product line currently only covers the IC layout stage. A complete digital design flow requires connecting all stages, and they haven't even finished building the framework yet. In other words, from the first line of RTL code to the final GDSII tape-out file, domestic chip design companies are running American software at every step.

Everyone in this industry chain is in the same ecosystem: equipment suppliers, contract manufacturers, IP licensors, design tool providers. The screws of the entire chain are all screwed onto Synaptics and Cadence.

For example, the testing equipment we just saw from KLA-Tex only supports Synaptics and Cadence's data reporting system.

Halfway there, at a spot on the side, Chen Yiran saw Chinese faces again, this time a team made up of several companies.

Several banners proclaimed the promotion of China's semiconductor industry. The booth was small, with only a few display boards.

The exhibition showcased the production capacity milestones of several companies, including SMIC, Huahong, and JCET. It appears this booth was rented by the Chinese government for these companies to attract more business.

Chen Yiran didn't go that way; he had more important things to do.

Since we're already here, we should at least meet with the people from Samsung and TSMC.