In the past decade, the rise of smart phones and the Internet has brought about tremendous changes in society, so society's expectations for semiconductor technology are increasing.

So how do you deal with the future development of semiconductors? Chen Ping, deputy general manager of TSMC China Business Development, proposed three magic weapons:

1. If you want to improve technology, you must continue the development of Moore's Law.

Chen Ping explained that the so-called Moore's Law is that when the price is constant, the number of components that can be accommodated on an integrated circuit will double every 18-24 months, and the performance will double. It is a very radical theory, but from 3μm in 1987 to 7nm, which has already started mass production, everyone seems to agree to follow and chase Moore's Law.

"In terms of TSMC, the 7nm mass production has been more than one year. 5nm has entered the initial mass production stage. Mass production will be realized by the end of next year. 3nm is already on the way. It is worth mentioning that our 2nm R&D is also It has already begun. We are constantly chasing Moore's Law, and it is still moving forward, and it has not failed." Chen Ping said.

What is the use of continuous improvement and reduction? Chen Ping gave such an example. Everyone knows that most of Huawei's mobile phones use their own chips. Huawei mate20 is a relatively hot mobile phone. This phone is equipped with Huawei Kirin 980 chip. The chip integrates 6.9 billion transistors, has very powerful performance and very low power consumption, all of which are implemented in the 7nm process.

The miniaturization of the process has benefited from the continuous innovation of the whole industry. The remarks that Moore's Law came to an end have been mentioned in 1992, but it has been continuing for so many years. It is too early to say that it is sung, Chen Ping said.

The lithography machine is a very important part of chip manufacturing. The lithography machine paused for many years when it developed 193nm, and finally achieved a breakthrough by immersion technology. The 7nm technology used today relies on a 193nm lithography machine. Nowadays, the lithography equipment company has made a major breakthrough. The euv technology makes lithography not become the bottleneck of miniaturization, and the new material has also made a breakthrough. This is why Chen Ping has confidence to say that Moore's Law will not end.

2. Introducing a large number of 3D integration concepts

Although the current technology is constantly improving, there are more and more requirements for end products, which require high-speed logic chips, memory, RF chips, etc. It is obviously inappropriate to spread the cake on a flat surface in the past. The high power consumption and large area of ​​the chip are completely opposite to the goal pursued.

Therefore, Chen Ping proposed that the current method is to replace the integrated circuit board with a semiconductor wafer board, and the wiring below, the different chips are connected below. If the chips are placed in parallel, this method is called a 2.5D system, if placed vertically It is called a 3D system. TSMC has already mass-produced 2.5D systems, and 3D systems are also being developed. In this direction, the development speed is parallel to Moore's Law.

As an example, Chen Ping mentioned the CoWoS technology, which is representative of the 2.5D system. Originally because of the high price, it was abandoned. As the process progressed to 16nm FinFET, and the heterogeneous chip integration trend was formed, many manufacturers have ordered it. The 2.5D system can provide high-speed computing, which is a relatively common technology. 

Chen Ping also mentioned another trend - advanced packaging technology, which combines different processes through heterogeneous integration, and the chips manufactured by completely different processes are put together to save functions and reduce the volume as much as possible.

3. Collaborative optimization of hardware and software

For the production of the product, the hardware is very important, and the software cannot be ignored. Chen Ping said that in the early development process, the mechanical method was adopted. Now it is different. You need to see if the final design is the optimal solution. Therefore, it is necessary to work closely with customers during process development, and not to use indicators as the ultimate goal.

At the system level, the advantages of the previous gpu, cpu and so on are more general and programmable. But the shortcomings are also obvious, the efficiency is relatively low, and with hardware acceleration, the functions are specific and the flexibility is not good. Today's designs turn hardware accelerators into high-speed devices with programmable, design-capable features that are at the system level.

As far as Chen Ping is concerned, the structure of the future soc will basically be optimized by hardware and software. TSMC will also pay close attention to the optimization of the system segment.