How did the S9 become the king of computers

Once upon a time: the founders reveal how the bitland S9 became the "king of machines"

Wu said blockchain

With the grand force mining machine collective admission, a generation of machine emperor ant mining machine S9 will withdraw from the historical stage. Reviewing the history, the author believes that the chengdu team achieved a cross-generational advantage over competitors through dynamic triggers, which was also the main reason for the success of S9. The design of S9 is the miracle of bitcoin mining machine. It has become the most productive mining machine in the bitcoin mining machine. With a life cycle of more than three years, it is called the machine king.

(the writer of this article is xie Dan, former technical director of bitmainland, and wu said that blockchain is authorized to forward. The views of this article are the author's personal opinions.)

Following is an article by xie Dan, former technical director of bitmainland, via wu shuo blockchain:

In August 2014, I opened a back-end design service company in chengdu. I want to provide value-added back-end design services. While looking for a client, I searched the Internet and decided that the company building the bitcoin mining machine would be in demand. & have spent

At that time, I wrote to many IC design companies, and bitland was the fastest to reply. Later, when chatting, I learned that this mailbox was used by wu jihan, and bitmainland was also looking for a plan to improve chip competitiveness. So wu jihan forwarded this email to zhan, who met me in chengdu in September. Jayne and I had two interviews and it was a good conversation. Zhan always felt that this project took a long time, was risky and had intellectual property rights concerns, so he suggested that I dissolve the company and lead the team to join bitmainland.

In October 2014, I officially joined bitmainland with a small team of two people and became the full customization department of bitmainland. The initial direction was the domino logic in the email.

Domino logic is a relatively mature dynamic logic structure, whose main form is as follows:

Compared with the corresponding static circuit, domino logic circuit has the following advantages and disadvantages:

1. Since the signal only needs to drive N tube, the capacitance is small and the speed is fast.

2. Since point A is A dynamic point, the final inverter M3/M4 is necessary.

3. There is A problem of charge-sharing at point A.

4, M1, M2, M3, M4 are newly added devices, and PMOS are reduced devices

5. Since M1 and M2 are driven by clock signals, their power consumption is at least twice that of other ordinary signals.

Therefore, in terms of power consumption, since M1 and M2 are clock signals, their power consumption is calculated at least according to the power consumption of two signals, then the power consumption of M1+M2+M3+M4 is increased by 6 MOS transistors. For general logic, there are not many standard cell libraries with more than 12 MOS transistors. In terms of area, M2+M4 increases the area of two NMOS tubes, while PMOS is less, but it is difficult to show the saved area in the layout. In other words: domino logic will speed up the chip, but the area and power consumption will increase slightly.

When we submitted the conclusion in January 2015, it was basically confirmed that domino logic was not suitable for mining machine chips. Because of the pure parallelism of the bitcoin mining chip, speed is not so important. The biggest cost of a mining machine is electricity, so power consumption is the most important. When we evaluate the rule of thumb for area and power consumption, power consumption is weighted more than three times as much as area and speed.

The failure of our domino logic attempt did not dampen our exploration, because we found a standard unit suitable for dynamic logic with a large volume and many pipes, and a clear example of such a standard unit is the flip-flop. So we go back to the beginning, with the dynamic trigger as the target.

At the beginning of integrated circuits, in the 1970s, because of the high cost of each transistor, the triggers at that time were all dynamic, whether it was domino flip-flop, C2MOS edge-triggered flip-flop, or TSPC positive-edge flip-flop. We found treasure in it.

For example, a TSPC positive-edge flip-flop has the following logic:

Still, if we count clock gate as two gates, this flip-flop is 4 times 2 plus 7 is equal to 15 gates.

The structure of our most common static trigger before was this:

Plus the inverse of the CLK, which we add together is & NBSP; 8 times 2 plus 12 plus 4 is equal to 32 gates. The flip-flop equivalent tubes in static logic consume more than 1 times more power than in dynamic logic.

Similarly, in terms of area, dynamic logic has 11 devices, while static logic has 22 devices, which is exactly double the area.

After we have identified the triggers for adopting dynamic logic, the next step is how to integrate them into our design process. Finally, we added a partial time constraint to the function description of the trigger with static logic, so as to prevent the leakage current in the dynamic logic from causing leakage at the dynamic capacitive point. In addition, some parameters of the static trigger are applied when timing and power consumption are extracted. In short, we added an outer box to the dynamic logic to make it look like a normal static trigger to the front-end designer, with no difference between front-end design and integration.

Having completed the library of dynamic triggers, we have a new trigger with half the original area, half the original power consumption and almost identical timing. Since bitcoins need to be counted all the time, there is no need for static triggers to hold data for a long time. When we have completed the dynamic triggers, the distributed operation and the complete pipelined logic of bitcoin are particularly suitable for dynamic triggers. For a pipeline line, the structure looks like this:

We can directly change it to:

Considering that the area and power consumption of the new dynamic trigger is only half that of the original one, the new pipeline doubles the speed with the same area and power consumption. In our bitcoin mining chip, we changed from a 32-level pipeline to a 64-level pipeline, doubling the computing power.

We completed the design of BM1385 chip at 28nm (ant miner S7) in mid-2015, and completed the design of BM1387 chip at 16nm (ant miner S9) at the end of 2015. In terms of performance, our 28nm is almost identical to the 16nm of our competitors, and our 16nm chip is half the cost of our competitors.With dynamic triggers, we achieve a cross-generational advantage over our competitors. In particular, the design of S9 is the miracle of the bitcoin mining machine. It has become the most productive mining machine in the bitcoin mining machine, with a life cycle of more than three years and is known as the machine king.

In the digital currency mining industry, which requires little software, a product costs half as much as a competitor, which is a huge competitive advantage, meaning you can have a price war advantage at your will. Because you don't make any money selling to your competitors, and bitland still has a gross margin of more than 50%. It is with the secret weapon of dynamic logic that the success of S7 and S9 has turned bitmainland, which had been fighting for less than 20% of the market, into a dominant player (more than 70%).

A direct result of the rise of bitmainland is the exit of foreign bitcoin chip companies. In 2014 and 2015, high-profile companies KNC, bitfury, Spondoolies-Tech, 21 Inc., all quickly declared bankruptcy or withdrew from the mining chip market.

With the departure of some bitland employees, the dynamic trigger technology gradually spread to other chip developers in the country, but the technology remained largely domestic. In 2017, Japanese GMO was still trying to enter the field at 12nm and 7nm. Judging from the company's publicity, they still adopted the static trigger structure. In addition, they encountered the bear market cycle in 2018, and then withdrew from the market at a loss a year later, which is expected.

Before 16nm, the new generation of process mask was within millions of dollars, and the technical, human and risk of redesigning dynamic logic were obviously higher than millions of dollars, so the advantages of dynamic logic could not be reflected. After 16nm, however, the new 10/7nm, and future 5nm, cost tens of millions of dollars. Dynamic logic can exceed the performance of a process node will make it appear more advantageous, with a new vitality. I look forward to more companies applying dynamic logic to bring this ancient art of design back to our times. & have spent

Original address:https://mp.weixin.qq.com/s/MGA2AGsJ8u0LKHVauJEeuQ

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia