Chapter 86: 20nm
For the drill, Li Qingsong intentionally delayed his commands, creating a communication lag of about 100 seconds between himself and both fleets. However, one fleet had no combat AI assistance, while the other did.
And the combat AI was deployed near the battlefield, so it had no such delay.
The moment the two sides engaged, Li Qingsong noticed something was wrong.
’Is this artificial intelligence? More like artificial stupidity!’
A warship in the thick of the fight clearly should have pushed forward to provide suppressive fire, creating an opportunity for its allies to launch their Interstellar Missiles.
But within that 100-second communication delay, before he could issue his next command, the combat AI preemptively took control of the warship and made it fall back in a panicked retreat.
It did manage to dodge an enemy Interstellar Missile, true, but in doing so it tore open its own defensive line and shattered the battle plan. This allowed the enemy to seize the opportunity, unleashing a volley of millions of rounds and inflicting heavy losses.
Another time, the fleet was supposed to launch an offensive, but under the combat AI’s control, its warships just sat there, motionless. They completely missed the tactical window.
These were the major blunders. The minor errors were even more numerous—simply countless.
In the end, the fleet without combat AI assistance—the one where Li Qingsong had to endure the 100-second communication delay—won a resounding victory, completely "annihilating" the side with AI support.
’More than ten years, over a hundred million tons of fuel, and trillions of rounds spent... all for this?’
Li Qingsong simply couldn’t accept it.
’This shouldn’t be happening. How could it turn out like this?’
Li Qingsong recalled the freighter carrying the supercomputer from the battlefield. He retrieved the combat AI’s operational data and assigned several thousand Clones to analyze its code execution and decision-making logic.
Through this analysis, Li Qingsong gradually discovered the root of the problem.
’It seems I was wrong to blame the combat AI...’
He had underestimated the difficulty of controlling warships in real-time combat, as well as the sheer volume of data that needed to be analyzed and processed.
It wasn’t that the AI was stupid. Rather, the moment the battle began, the volume of data and the computational load were so immense that the supercomputer overloaded. It couldn’t keep up, resulting in massive lag, calculation errors, and logic failures that ultimately led to this situation.
Li Qingsong breathed a sigh of relief.
’Thankfully, it’s not a problem with the combat AI itself. It seems the last ten-plus years of investment weren’t wasted after all...’
Now that he had identified the problem, the solution was straightforward.
If the supercomputer’s computational power was insufficient, then he would just have to develop one with even more!
With parallel algorithms and architecture already pushed to their limits, there was only one path left to create a supercomputer with greater computational power.
Develop more advanced chips!
Currently, Li Qingsong had already mastered the mass production of 45-nanometer process chips.
Following the technological roadmap already proven by the Human World, Li Qingsong’s next step should have been to develop 28-nanometer process chips.
However, factoring in his continuous research and technological progress over the years, Li Qingsong decided to bypass 28nm chips and jump directly to the next generation: 20-nanometer process chips!
Once 20nm chips were successfully developed, the number of transistors on a single chip would skyrocket from approximately 1 billion to 4 billion.
The overall performance of a single chip would also increase by 60% to 100%.
But that wasn’t the most important part. Critically, chips with a more advanced process could accommodate more sophisticated parallel algorithms and supporting hardware. This meant Li Qingsong could use more chips in parallel within a single supercomputer, potentially boosting its computational power by a factor of 40 over the current baseline!
The prospect was incredible, but jumping directly from 45nm to 20nm while skipping 28nm would clearly present enormous obstacles and extreme difficulty.
At the computational technology research base, Li Qingsong once again assigned numerous Clones to build upon their existing foundation and begin research into the more advanced chip.
After a brief period of preliminary research, the technical hurdles he needed to overcome were organized into several major categories and presented to him.
’First, the deep ultraviolet light source I’m currently using has too long a wavelength. Its resolution has hit the physical limit. The diffraction effect is blurring the patterns, making it impossible to etch smaller transistor structures.
Second, with traditional transistors at this scale, the short-channel effect is pronounced, making power consumption and performance difficult to control.
Also, the physical properties of the traditional silicon gate dielectric are no longer sufficient. That needs to be improved as well...’
Li Qingsong divided the Clones into several large teams, each tasked with tackling one of these key challenges.
After tireless, round-the-clock research, the first breakthrough came in the light source.
Li Qingsong developed an extreme ultraviolet (EUV) light source with a much shorter wavelength, reducing it to just 13.5 nanometers and thereby dramatically increasing the resolution.
In addition, Li Qingsong modified the existing process, switching from single to multiple exposures. This increase in process steps enabled a massive improvement in fabrication precision.
Following that, Li Qingsong found a material with better performance, using a metal gate to replace the silicon dioxide gate and achieving yet another technological breakthrough.
Six months had now passed. Li Qingsong had conquered most of the other technical hurdles, but the most difficult and crucial obstacle still lay before him.
The short-channel effect.
When a transistor’s channel length is shortened beyond a certain point, the laws of physics cause its electric field distribution and carrier behavior to change. This leads to a series of problems, such as a drop in the transistor’s threshold voltage and drain-induced barrier lowering, which severely affect the chip’s performance and reliability.
To solve this problem, Li Qingsong had already tested thousands of solutions—including optimizing transistor layout and design, enhancing material properties, and refining production processes—but all ultimately proved ineffective.
’I’ve solved every other problem, but I just can’t crack this one. It’s just so...’
Li Qingsong sighed.
By now, the time and resources consumed by the 20nm chip research had far exceeded projections, affecting Li Qingsong’s overall scientific agenda.
But there was no alternative. Giving up halfway was out of the question. The 20nm chip was simply too important to Li Qingsong. It wasn’t just for the combat AI; his other AIs—for intelligent factory automation, automated starship control, and even research in fundamental physics, mathematics, chemistry, and engineering—all relied on supercomputers with greater computational power.
On this particular day, Li Qingsong still had tens of thousands of Clones conducting experiments, and had devoted the brainpower of several thousand more specifically to contemplating this single problem.
It was in this state that an idea suddenly formed in the mind of a particular Clone and was transmitted to Li Qingsong’s consciousness.
’Since planar transistors cause the short-channel effect... could we change the transistor from a two-dimensional structure to a three-dimensional one?’