Rebirth: Starting from Lighting Up the Tech Tree

Chapter 236 Intelligent Manufacturing

After the mass production of Interstellar Neuron, supply chain bottlenecks were exposed.

The NX series chips are the heart of the entire brain-computer interface system. Each chip is manufactured using a 4-nanometer process and integrates over 30 billion transistors, responsible for all the computations involved in collecting, compressing, and transmitting brain signals. The design of this chip was completed by Fang Ze's team, while manufacturing was outsourced to Samsung's foundry. The process parameters underwent multiple rounds of optimization in the early stages of mass production, but the yield rate remained stuck at a certain number.

73 percent.

Yield means that out of every 100 wafers produced, only 73 chips pass functional testing. The remaining 27 chips either have micro-defects during manufacturing or fail to meet performance standards in the final packaging stage.

Fang Ze provided a straightforward calculation at an internal meeting: "Every percentage point increase in yield means more than 5,000 usable wafers per batch. The market value of 5,000 NX-40 wafers exceeds 100 million RMB. One percentage point is 100 million."

Samsung's foundry team has already done three rounds of process optimization. The photoresist formula has been adjusted twice, the etching gas ratio has been changed three times, and the ion implantation energy parameters have been tried no less than twenty combinations within the allowable range. The yield fluctuates slightly after each adjustment, but it has never broken through the ceiling of 73%.

Shen Yiming proposed a completely different approach at a technical discussion meeting.

"We've been using trial and error to optimize process parameters," he said, looking at Fang Ze. "Samsung's engineers are very capable, but semiconductor manufacturing has too many variables: photoresist thickness, etching depth, ion implantation angle, annealing temperature—each dimension has dozens of possible combinations. The human brain can't exhaust all the combinations, but quantum computing can."

His solution is to abstract the entire manufacturing process of the NX-40 chip into a high-dimensional optimization problem, and then let the Tianyan quantum computer perform a global parameter space search. Instead of having engineers try each option one by one, the quantum computer calculates all possible parameter combinations at once and finds the globally optimal set.

After listening, Fang Ze thought for five minutes and then asked a question: "Can quantum computing simulate semiconductor manufacturing processes? The materials science involved in the manufacturing process is extremely complex. Is the accuracy of the quantum computing model sufficient?"

Shen Yiming's answer was: "Perfect simulation isn't needed. We're doing parameter sensitivity analysis. We don't need to know the complete mechanism of every physical process; we only need to know which parameter changes have the greatest impact on yield. We calculate the parameters with the greatest impact using high-precision models, and use statistical approximations for the secondary parameters, then run a mixed analysis."

Samsung is very positive about this solution. Their process engineering team has already exhausted all the adjustments that can be made using traditional methods in the previous three rounds of optimization. If they want to continue to improve yield, they do need a completely new tool.

The Tianyan quantum computer completed the parameter space search, which the Samsung team had estimated would take two years to complete, in just three days.

Three days versus two years. This comparison is newsworthy in itself. But the real news lies in the result.

Tianyan provides the optimal parameter combination for seven key variables. The photoresist thickness tolerance is narrowed from ±0.5 nanometers to ±0.1 nanometers, the etching depth compensation value is adjusted from the usual 2% to 3.7%, and the ion implantation angle deviation calibration is corrected from 0.5 degrees to -0.3 degrees. The remaining four parameters involve the annealing temperature profile, cleaning solution concentration, photomask alignment offset, and wafer carrier cooling rate.

Each of the seven parameters has a small adjustment range, with the largest change being only a few percentage points. However, the combined effect of the seven parameters is astonishing.

Samsung adjusted its production line according to the parameters provided by Tianyan. The adjustment itself only took two days because the changes to the parameters were all within the controllable range of the existing process window. Trial production began on the third day.

Yield: 91 percent.

When Fang Ze saw this number in the production line test report, he called Shen Yiming and said, "You did in three days the work that Samsung engineers did in two years."

Shen Yiming replied even more bluntly: "I didn't do it, Tian Yan did it. I just gave it a challenge."

The vice president of Samsung's semiconductor division sent an email from South Korea to headquarters 402 that day. The email contained only one sentence.

Your quantum computer should be renamed Brain Making.

Zuo Cheng smiled when he saw the email, but quickly stopped. He was thinking about a bigger problem.

Quantum computing's application to manufacturing optimization isn't limited to chips. Any industry requiring high-precision manufacturing faces the same dilemma: there are too many dimensions of process parameters for the human brain to exhaustively enumerate them. Semiconductors are an extreme example, but the forging process of aero-engine blades, the welding parameters of new energy vehicle battery modules, and the biocompatible coating ratio of brain-computer interface electrode arrays are essentially isomorphic problems.

He shared this idea with Shen Yiming. Shen Yiming's response was to immediately begin writing a generalized manufacturing optimization module.

Within a month, quantum manufacturing optimization solutions expanded from chips to four new areas. The pass rate for aero-engine blades increased by 15 percentage points. The consistency issue in new energy vehicle battery modules decreased from a 7% defect rate to 1.2%. The biocompatibility test pass rate for brain-computer interface electrode arrays increased from 88% to 99%. A joint venture manufacturing high-end medical devices used the quantum optimization engine to solve the heat treatment cracking problem in precision stainless steel parts, reducing scrap losses by over 80 million annually.

More than 40 manufacturing companies across seven industries have connected to the Tianyan Quantum Cloud Manufacturing Optimization Module.

The average yield rate has increased by more than 12 percentage points. This translates to annual cost savings of over 30 billion RMB.

When the news reached the Ministry of Industry and Information Technology, a vice minister in charge of manufacturing remarked at an internal meeting: "Quantum computing is not a matter for the laboratory, but a matter for the workshop."

That evening, Zuo Cheng sat alone in his office, reviewing the summary report of all the day's data. The breadth of the manufacturing sector covered exceeded his initial expectations. He opened the system panel.

In the civilization perception interface, the color of the light pillar representing the manufacturing sector changed from light blue to dark blue. The thickness of the pillar didn't change much on a visual scale, but the data density within it significantly increased. He noticed new connecting lines appearing between the manufacturing pillar and several other pillars. Traffic data is feeding back into the AI ​​model's training accuracy; the improved AI model accuracy makes the manufacturing optimization module's search efficiency higher; the cost reduction brought about by manufacturing optimization makes more companies willing to connect to the quantum cloud; and the increased data volume on the quantum cloud, in turn, enhances optimization capabilities in all directions.

The flywheels in six directions are meshing.

Before closing the panel, Zuo Cheng glanced at the location of the ninth branch. The gray outline was still there, a little brighter than when he had looked at it before.

He compiled the data report on manufacturing optimization and sent it to Han Lu, asking her to submit it to the Ministry of Industry and Information Technology the next day.

On the same day that the manufacturing optimization report was submitted to the Ministry of Industry and Information Technology, Han Lu received a new email. The sender was a vice minister from the Ministry of Agriculture.

The email subject was a single sentence: "Quantum brains can manage manufacturing, but can they manage farmland?"