[New Zhiyuan Introduction] Phillip Wong, head of R&D at TSMC , said at a Hot Chips conference held in Silicon Valley recently that Moore's Law is not dead, it is still effective and in good condition. He also stated that with the advancement of new technologies, transistors will reach 0.1 nanometers by 2050, which is about the size of hydrogen atoms.
"There is no doubt that Moore's Law is still effective and in good condition. It has not died, slowed down, and got sick."
TSMC's R&D head Philip Wong (Huang Hansen) said at the Hot chips conference that he showed TSMC's 's prospects for chip technology, saying that by by 2050, the transistor will be reduced to the hydrogen atomic scale, that is, 0.1nm.
Hansen Huang began to serve as vice president of corporate research at TSMC in August last year. Before that, he was a professor of the Department of Electrical Engineering at Stanford University. He is good at researching new forms of storage technology and has a high reputation in the academic community.
Hansen Huang's speech at the Hot Chips conference was titled: "What will the next semiconductor process node in bring to us? ". He also introduced the development of other chip technologies in detail, such as placing memory directly above of processor, which he expected to improve performance.
is very optimistic about Moore's law. TSMC proposes three major directions for chip progress
Moore's law predicts that the number of transistors that can be accommodated on an integrated circuit will double every 18 months, and the performance will also double (that is, more transistors make it faster).
However, how to put billions of transistors in a chip in the most economical way has become the biggest challenge in chip manufacturing at present, so in recent years, many people believe that Moore's Law is approaching the physical limit and has begun to slow down.
Intel has been working hard to develop advanced processes, but from the perspective of the entire industry, the price of individual transistors will no longer continue to decline. This limits that new manufacturing processes can only be used for high-end, high-cost chips. Gone are the good days of the chip industry in the past, when the chip's clock speed increased, but its power consumption was not affected at all.
Therefore, it is not surprising that pessimists appear in the chip manufacturing industry.
Huang Hansen expects that the processor will be composed of 3D stacks of different chip elements, which are usually separated at present. This will mean that the chip gets smaller size and higher performance.
However, , as the leader in wafer foundry, TSMC is very optimistic. Huang Hansen said Moore's Law is progressing well and boldly predicted progress by 2050, although he did not provide any detailed plans. David Kanter, an analyst at
Real World Technologies, was more cautious. Since TSMC is now tied with Intel , rather than after Intel , TSMC has to assume more leadership responsibilities and increase R&D investment, it is not surprising to hear that the company is so optimistic. But when it comes to chip advancements, Huang Hansen shys away from some practical issues, such as the slowdown in shrinking transistors and the increased cost of manufacturing the latest generation of products.
fundamentally improves
"We look forward to seeing more innovations in different directions, which will provide sustained benefits for the industry." Huang Hansen said: "That's what we care about."
Huang Hansen said that the components of chip technology are shrinking to extremely small sizes
Regarding the future technology route, Philip Wong believes that like carbon nanotubes (1.2nm Scale), 2D layered material , etc. can make transistors faster and smaller in size; at the same time, phase change memory (PRAM), rotation torque transfer random access memory (STT-RAM), etc. will be directly packaged with the processor, reducing the volume and speeding up data transfer; in addition, there is 3D stacking packaging technology .
Specifically, Huang Hansen put forward several views on the future development direction: the new technology of
- will make transistors faster and smaller.A technology that has long been considering—carbon nanotubes—is now becoming practical. Another is a material called a 2D layered material that can provide similar enhancement by allowing electrons to flow through the chip more easily.
- Some new memory technology will be built directly into the processor , rather than as a separate chip connection. This quick connection will greatly improve performance because the logic circuit on the chip (the part that processes the data) will get the required data faster, so there is no need to have too much idle time.
- 3D stacking technology will mean that the current isolated computer processor functions can be sandwiched in multiple layers and connected to high-speed data paths.
"In these systems, multi-layer logic and memory are integrated in a fine-grained way, and connectivity is key," said Huang Hansen.
analyst Nathan Brookwood said that although Huang Hansen is very concerned about technologies such as carbon nanotubes, he does not believe that TSMC itself will bet on any specific new technology at this stage.
However, Huang Hansen emphasized that in addition to hardware, software algorithms also need to catch up. Once this is achieved, advances in chips will provide better computing devices. This is crucial, Huang Hansen said: " society's demand for advanced technology is endless ."
Next, Xin Zhiyuan brought the complete PPT of Huang Hansen's keynote speech at Hot Chip 2019, with detailed interpretation.
TSMC Hot Chips Conference Speech Compilation (with PPT)
Moore's Law talks about component density, which is the main driving force of high-performance computing.
From the logarithmic graph, Moore's Law not only does not die, but also lives well. The density of transistors is still increasing, and it will continue to increase in the foreseeable future. As for new attributes such as clock speed and operating efficiency, which people are also concerned about, actually exceed the scope of Moore's Law.
Entered the era of AI and 5G, and the "memory wall" problem is becoming increasingly prominent. The demand for the flow and transfer of massive data in is getting higher and higher, and memory access determines the energy efficiency of computing.
Deep neural network requires a large amount of memory capacity, and the problem of memory shortage will be more prominent in the future. More amounts of SRAM are needed on the chip, but it is never enough. What is important in is what kind of memory.
In the existing systems, most of them are 2D and 2.5D, using TSV, we need to take a step further and enter 3D.
The next step is Beyond 3D, which implements multi-layer integration of logic and memory, and implements a high-density TSV process on the nanoscale scale, namely the "N3XT-level" system.
The next generation of memory needs to have high bandwidth, high capacity, and it needs to be on chip.
Research shows that memory with the above conditions can increase system-level returns by nearly 2,000 times. Of course, it is difficult to achieve with the existing technology. It is difficult to build high-performance transistors on the upper layer because the memory layer will melt when manufacturing requires 1000 degrees of high temperature conditions.
To implement the ideal system mentioned above, ultra-thin equipment layers and low manufacturing temperatures are required.
In recent years, transistor technology has achieved a lot of progress, with over-metal design of 2D layer materials, 1D carbon nanotube design, etc. These materials are very light and thin, greatly reducing the channel width of the transistor, but still maintaining a high mobility level.
realizes the integration of memory and logic platforms under the 3D architecture, making the advancement of transistors and manufacturing technology a continuous unity.
. To achieve this goal, it is not possible to fight alone. This requires close cooperation between system engineers and developers, closer exchanges in hardware equipment manufacturing technology and needs, and closer ties between the academic and industry.
