Before May Holiday, by the Huangpu River, a gentle spring breeze caresses the face.
The sun lazily shines on the window, while the singer on Riverside sings love songs with great enthusiasm. I anxiously sit on a leather chair at Starbucks, staring at the Flatwhite in front of me.
“Are you Mr. Jin?” The voice was as graceful as a startled swan.
Emma is a technical expert who has just returned from 2035. She is wearing a gray T-shirt with no words on it, and her hair hangs naturally on her shoulders without any sign of fatigue.
“Do you remember our agreement? We cannot discuss what will happen in the future, only talk about logic.” Emma said.
I nodded and opened the prepared outline of questions on my phone.
Q&A Session: Link to heading
Jin Jiefan: It is now the year 2021, and there is much uncertainty regarding the development of Moore’s Law. Can you speak to this?
Emma: Human understanding of physical limits is still very naive, and the idea of defining an end point for Moore’s Law is too narrow-minded. Furthermore, the current definitions of 10nm, 7nm, and 5nm are not related to physical sizes, as they simply rely on multiplying by a rough factor of 0.7 each time.
Emma: Moore’s Law has never been about line width, it’s about density. If we calculate the number of transistors per unit area of a packaged chip, it’s easy to scale up by many magnitudes. Just look at 3D NAND today. Logic chips can also be made into 3D, it’s just that your technology is lagging behind.
Jin Jiefan: There is a theory called “death economics”, which means that when the monetary threshold for technology becomes too high, it becomes impossible to continue.
Emma: The information revolution for humanity has actually just begun. You cannot even imagine how great your future demand for computing power and storage will be.
Emma: The 400 million taels of silver paid in the Xinchou Treaty of the Qing Dynasty wasn’t even enough to build the “Three-piece Suit” in Lujiazui. Money is not something to worry about.
Jin Jiefan: Alright, let me ask some practical questions. Currently, carbon nanotubes have shown promising results in laboratory experiments. Will they become the backbone of technology below 3nm?
Emma: I cannot tell you the result. Carbon nanotubes require separate preparation, such as rolling graphene into small tubes and precisely depositing them onto the silicon wafer after lithography. Your most advanced atomic layer deposition High K materials are not suitable for carbon nanotubes. You also need vast amounts of new technological equipment, and I’m sure you’re tired of the Damascus process after decades.
Jin Jiefan: Understood, it seems like this path is too difficult. Maybe a 3D SoC would be easier?
Emma: At present, there are significant problems with the balance in human chip technology. Mainstream memory has not made much progress in decades, and the von Neumann architecture requires memory and computing to be separated. DRAM’s bandwidth, speed, and power consumption are lagging behind. Doubling computing power may not necessarily require stacking transistors; architecture is a core point.
Emma: The computing power and storage of the human brain originate from the solar energy obtained through plant and animal food sources. Just think, by simply eating a few slices of bread a day, the real-time image processing ability of the human brain surpasses even the most powerful 200 TOPS chips.
Jin Jiefan: Alas, everyone knows the issue with DRAM, but every time it manages to defeat all competitors. Now, the power consumption of DRAM has exceeded that of the CPU in many systems, yet its speed remains slow.
Emma: 3D SoCs such as HBM can solve some bandwidth and chip size problems, and maybe you can solve the troubles of heat dissipation and cost in the mid-term. However, in the long run, the bottleneck of memory speed is still a problem that humans need to solve.
Emma: The amount of data generated by humans and various IoT devices every day will quickly surpass your existing processing capacity. Machine learning and neural networks will inevitably drive the application of in-memory computing and new forms of memory.
Jin Jiefan: Does new memory refer to MRAM, RRAM, and PCRAM, etc.? I know that due to its advantages in read and write speed and power consumption, FRAM has been widely introduced in the IoT field, such as electric meters.
Emma: I cannot tell you the outcome or state that DRAM will be replaced, as there is still much to be explored. Additionally, the 3D SoC encompasses integrated components such as the GPU, NPU, and FPGA, among others, but humans have not yet fully understood the relationship between processing power and architecture.
Jin Jiefan: It’s no wonder that Nvidia wants to acquire ARM, AMD wants to acquire Xilinx, and Intel has acquired Altera. They are probably aiming to launch a massive XPU battle in the data center.
Emma: Not just data centers, but also the so-called edge computing demands are significant. The electron flow generated from data consumption consumes a substantial amount of energy, mostly in networks, motherboards, and even in buses.
Emma: You will eventually understand that traditional general-purpose CPUs and von Neumann architectures come at a high cost. However, your programmers have yet to realize what the new programming methods should be under the new compound architecture pattern.
Jin Jie Fan: Hmm, it sounds like the field of computer science still has a promising future for the next few decades. Let’s go back to the stage of semiconductor manufacturing. The High NA EUV lithography machine will soon be produced, but I think it would be more cost-effective to use multiple EUV exposures first, wouldn’t it?
Emma: Your logic makes sense. In terms of technology, it is more reasonable for low-NA EUV to have two or four exposures. However, if we start using high-NA at 2nm, there are many accompanying technical issues that need to be solved synchronously, such as photoresist, mask, and metal deposition.
Emma: Also, actually NA=0.55 should be called Middle NA and NA=0.75 should be called High NA. It’s just that the depth of field will be shorter and the precision of wafer stage and mask movements will be higher.
Jin Jiefan: It seems that studying mechanical engineering can also bring youthfulness. I heard that ASML regrets wasting photons during the start of the movement of the wafer stage, as EUV radiation is extremely precious. In order to ensure productivity, they must race against time and increase the speed of the wafer’s movement as much as possible. However, there cannot be even a hint of vibration when the stage rapidly accelerates and decelerates.
Emma: Yes, this is admirable. Additionally, the aid of deep learning technology can help engineers compensate for and improve image quality, similar to how your phones can use computational photography and Photoshoping moon technology can be used to correct exposure lines.
Jin Jiefan: This is quite interesting, it seems that the bottleneck for Moore’s Law in the next ten years will not be on photolithography. I would like to ask another related question, what is the future of digital currency?
Emma: This question is a bit inappropriate to ask. All I can say is that when the amount of credit currency issued increases, it will naturally flow to different places. An excessive amount of currency does not necessarily cause significant inflation. It can flow into newly generated or newly assetized things. Different digital currencies, like different stocks or collectibles, have audiences across different social strata.
Emma: The highest predator in the financial food chain, always borrowing from others to make low-risk arbitrage. After the lower class solves the problem of food and clothing, they always try their luck and attempt to move up on the social ladder with any leftover money.
Emma: But they are just data sources in reality, and the upper class will always be the data processors. People’s savings will be lent to others, stocks or even digital currencies purchased will be borrowed by others, and they bear all the risks, but only a few can achieve success because most of the profits are taken by predators.
Jin Jiefan: This conclusion is quite unsettling. Our phones here always display fake red packets or urge users to choose small loans for payment. The creators of these apps have long become billionaires.
Jin Jiefan: Let me ask a more cheerful question. Recently, I have been playing Pico’s NEO and Oculus’s Quest2, and I really like them. Although I haven’t played games for a long time due to my age, it’s very easy to get started with VR now and every game is a masterpiece. The movie-watching experience is also impressive. Additionally, the immersive feeling is truly different from traditional gaming consoles, which makes it very exciting. Will VR become a huge industry in the future.
Emma: Do you want to find out if you should buy Facebook stocks? I can’t tell you the answer. VR can truly revolutionize the indoor lifestyle of humans, while AR can demonstrate its power in the industrial field. However, the key question you need to ask yourself is whether humans will become increasingly disinclined to go out.
Emma: After solving the bandwidth problems for 8K and 16K videos, there is no need for people to go to the cinema anymore. Just put on a pair of glasses and you can have the same visual and even sensory experience as in the cinema. After solving the bandwidth problems for 32K and 64K videos, even traveling is unnecessary. You can put on VR glasses and be wherever you want - in any country or location, even on Mars.
Jin Jiefan: I’m not sure why, but this doesn’t sound good. If people stop going out, will Tesla become the world’s best-selling car company?
Emma: I’m sorry, I can’t say. The meeting today is over, and I have already felt the disturbance in Hilbert Space.
Jin Jie Fan: I want to ask one last time: Can humanity overcome the novel coronavirus? Will the future world become better or worse?
Emma smiled and said, “Anyway, it will be different from what you imagine.”