The exercise method adopted by Li Qingsong was to intentionally delay the time of issuing orders, so that both fleets had a communication delay of about 100 seconds with themselves, but one side had no combat AI assistance and the other had combat AI assistance. And the combat AI is deployed near the battlefield without any delay. As soon as the two sides started fighting, Li Qingsong discovered something was wrong. Is this artificial intelligence? Is this artificial intelligence? Obviously, a warship should advance in a melee, provide fire suppression to friendly forces, and create conditions for them to launch interstellar missiles. But during the 100-second communication delay, before he could issue the next order, the combat AI took the lead in manipulating the warship to retreat directly and hastily. It was indeed good that it dodged an interstellar missile from the opposite side, but it directly tore through its own defense line and disrupted its own battle plan, which directly led to the opponent seizing the opportunity and launching a salvo of millions of bullets, causing heavy losses to its own side. Another time, this side should have launched an attack, but under the control of the combat AI, its warships just stood there motionless, ultimately leading to the missed opportunity. This is a relatively big mistake. There are many other smaller mistakes, too numerous to count. In the end, in this exercise, the side without combat AI assistance won a great victory while Li Qingsong endured about 100 seconds of communication delay, and "completely wiped out" the side with combat AI assistance.
"More than ten years, more than 100 million tons of fuel consumed, trillions of bullets consumed, and this is the result?" Li Qingsong simply couldn't accept it. "It shouldn't be like this? How could this happen?" Li Qingsong called back the cargo ship carrying the supercomputer from the battlefield, retrieved the computing data of the combat AI, and dispatched thousands of clones to begin analyzing its code operation and decision-making logic. After this analysis, Li Qingsong gradually discovered the problem. I seem to have wrongly blamed this combat AI... Because he underestimated the difficulty of real-time control of warships on the battlefield, as well as the amount of data that needed to be analyzed and calculated. It's not that this AI is mentally retarded, but at the beginning of the battle, the amount of data and the amount of calculation were too large, which resulted in the supercomputer being overloaded and unable to keep up with the calculations. There were a lot of delays, calculation errors, and logical errors, which ultimately led to this situation. Li Qingsong breathed a sigh of relief. "Fortunately, it's not a problem with the combat AI. It seems that the investment over the past decade has not been in vain..." Once we find the problem, everything will be easy. Since the computing power of the supercomputer is not enough, then we should develop a supercomputer with greater computing power! At a time when parallel algorithms and structures have almost reached their limits, there is only one way left to create a supercomputer with greater computing power. Develop more advanced chips! At present, Li Qingsong has mastered the large-scale manufacturing technology of 45-nanometer process chips. According to the technological route that has been practiced in the human world, Li Qingsong should develop 28-nanometer process chips for the next generation. But combined with the uninterrupted research and technological progress over the years, Li Qingsong decided to bypass the 28-nanometer chip and directly research the next-generation 20-nanometer process chip! Once the 20-nanometer chip is successfully developed, the number of transistors on a single chip will skyrocket from about 1 billion to 4 billion. The overall performance of a single chip will also be improved by approximately 60% to 100%.
But this is not the most important thing. The most important thing is that chips with higher process can be adapted to more advanced parallel algorithms and supporting hardware. In a single supercomputer, Li Qingsong can use more chips for parallel computing. The computing power of a supercomputer can be increased by about 40 times on the existing basis! The prospects are so bright, but upgrading directly from 45 nanometers to 20 nanometers, bypassing 28 nanometers, obviously means huge obstacles and extremely high difficulty. In the computing technology research base, Li Qingsong once again selected a large number of clones, and based on the original research, they began to study more advanced chips together. After a short period of preliminary research, the technical difficulties that he needed to overcome were divided into several major categories and presented to Li Qingsong. “First, the wavelength of the deep ultraviolet light source I am using now is too long, the resolution has reached its limit, and the diffraction effect causes the pattern to be blurred, making it impossible to write smaller transistor structures. Secondly, at this size, traditional transistors have obvious short channel effects, and power consumption and performance are difficult to control. In addition, the physical properties of traditional silicon gate dielectrics cannot meet the requirements, and this also needs to be improved..." Li Qingsong divided the clones into several large teams, each of which tackled difficult problems in different areas. During the non-stop research day and night, the first breakthrough was in light source. Li Qingsong developed an extreme ultraviolet light source with a shorter wavelength, shortening the wavelength of light to only about nanometers, thereby greatly improving the resolution. In addition, Li Qingsong also changed the previous process, changing the previous single exposure to multiple exposures, which greatly improved the processing accuracy by increasing the process flow. After that, Li Qingsong found materials with better performance and used metal gate instead of silicon dioxide gate, achieving another technological breakthrough. Half a year had passed. Most of the other technical problems had been solved by Li Qingsong, but the most difficult and most important technical obstacle still stood in front of him. Short channel effect. When the length of the transistor is shortened to below a certain extent, the electric field distribution and carrier behavior will change due to the laws of physics, which will lead to a series of problems such as a decrease in the threshold voltage of the transistor and a decrease in the drain barrier, which will seriously affect the performance and reliability of the chip. In order to solve this problem, Li Qingsong has tried thousands of solutions, including optimizing transistor distribution and design, improving material performance, improving production processes, etc., but they were ultimately proven to be ineffective. "I've solved all the other problems, but I can't solve this one. This is really..." Li Qingsong sighed. At this moment, the time and resources consumed by the 20-nanometer chip research institute have exceeded expectations and have affected Li Qingsong's overall scientific research plan. But there is no way to give up halfway. The 20-nanometer chip is too important to Li Qingsong. It is not only used in combat AI, but also in other AI, including intelligent factory equipment, intelligent spacecraft control, and even basic physics, mathematics, chemistry, and engineering research. All of them are inseparable from supercomputers with greater computing power. On this day, Li Qingsong was still controlling tens of thousands of clones to carry out related experiments, and he also mobilized thousands of additional brainpower to think specifically about this issue. In this situation, a thought suddenly emerged from the brain of a clone and was transmitted to Li Qingsong's consciousness. "Since planar transistors lead to short channel effects, then... can we change the transistor structure from two-dimensional to three-dimensional?"