What is the "dual-core stacking" technology?
In terms of chip design, Huawei has also newly exposed a "dual-core superposition" patent, which can even make 14nm chips comparable to 7nm performance after being optimized.
According to the patents exposed by Huawei, Huawei's "dual-core superposition" patent can indeed greatly improve the performance of 14nm chips. Of course, if it is to be compared with the real 7nm chip, there may still be a certain gap, just like Intel's 14nm++ process.
Specifically, the difference between 14nm and 7nm is that under the same chip area, 7nm can have more transistors, and the number of transistors limits the performance of the chip.
This means that a 14nm chip is no better than 7nm no matter how optimized it is, just like pouring two cups of 50° water together, it can't reach 100°. However, if the chip tasks are divided through a certain technology, the 7nm chip does it all by itself, and the two 14nm chips that are stacked together complete a part respectively, and then the final results are superimposed, then the task of the 7nm chip can also be completed. Simply put, it is to divide the tasks of the two chips and superimpose them together to form a stronger operating efficiency. In theory, it is possible to optimize 14nm to be comparable to 7nm, but it also needs to solve the problems of power consumption, signal synchronization, and data stream co-processing.
Of course, in this way, the power consumption is bound to increase a lot, and this is the difficulty of "dual-core superposition".
Many people in this patent understand that two independent chips are physically stacked to achieve performance breakthroughs. In fact, this is a very serious mistake. If you rely solely on physical stacking, there will be many disadvantages that cannot be solved, such as compatibility and stability. It is impossible to solve the problem through physical stacking, such as sex, heat control, and design ideas. It is meaningless to lose more than gain.
The dual-core stacking level is used in the initial stage of design and production, that is to say, in the design process, the original chip is designed into a double-layer chip and then using its own unique technology to encapsulate the two-layer chip in a single chip. Synchronous signal mode and some other methods can activate the dual-layer chip to work together to achieve a breakthrough in chip performance. Therefore, when a physical layer is stacked, the design thinking is changed at the beginning of a design. These are two completely different ways.
Intel has achieved similar technological breakthroughs in history, but Intel has completed a new design through physical packaging. From monolithic to 2D integration, and then from 2D integration to 3D integration, ultimately meeting the needs of different application scenarios. When AMD's dual-core processor Athlon 64 X2 took the lead in 2005, Intel encapsulated the two processor cores on a substrate through a packaging process and launched the Pentium D series. The dual-core of the Pentium D series has not changed much from the Pentium 4. It is a dual-core processor that encapsulates two Pentium 4 processors on a substrate, which is fundamentally different from the native dual-core of AMD Athlon 64 X2. Intel’s approach is due to the lack of shared memory and independent bus interconnection for the Pentium D’s dual-core processor, so the performance of the Pentium D is far inferior to AMD’s Athlon X2. Instead, the power consumption of the processor has soared until Two years later, after the launch of the Core 2 series of true dual-core processors, Pentium D quickly withdrew.
Intel came to the top in the dual-core processor competition and began to launch the tick-tock plan, relying on its own advanced process and processor core upgrades to gain a competitive advantage faster and faster, while AMD had to sell its chip manufacturing business due to a lack of funds. The core competition is still at a disadvantage, and the processor has reached eight cores by 2012.
In the face of Intel's competitive advantage, AMD's FX series has also adopted the packaging method to package two quad-core processors together to achieve an eight-core architecture when the multi-core technology research and development are lagging behind. However, although AMD's FX series has eight cores, it uses two cores to share a floating-point unit. However, the application software at the time still required a lot of floating-point operations. In order to further improve performance, AMD increased the CPU frequency to 5GHz. The result is soaring power consumption so the AMD processor sales end has to force the sale of water-cooled radiators when selling the FX9590 series.
It can be seen from the practices of Intel and AMD that packaging technology alone cannot achieve a result of 1+1 greater than 2. On the contrary, the result of this approach is that the power consumption is too high, which is not conducive to heat dissipation, and the performance improvement that can be obtained in actual use is far Less than expected.
In the final analysis, it is currently known that the domestic chip manufacturing process can only achieve a maximum of 14 nanometers. Although it is moving towards 7nm, it is because there is no state-of-the-art UV lithography machine, so even if SMIC can develop 7 nanometers, There is no way to conduct trial production. If the chip can be optimized through this dual-core superposition technology, so that the performance of the 14nm chip can reach the 7nm performance standard, it can be regarded as a different way for Huawei.