The hidden story of Antminer S9 revealed

01

If I were to recall the development process of the BM1387 chip, the following keywords would come to my mind first:

"Prepare for a rainy day"

As we all know, BM1387 uses TSMC's 16nm process and was put into production in January 2016. But what most people may not know is that we have started the research and development of 16nm since the first half of 2014. At that time, it was still a relatively early stage of 16nm, and only a few large semiconductor design companies had introduced the 16nm process. Bitmain's chip R&D team had just been established. Even in July 2014, when Bitmain first visited TSMC to discuss 16nm technology, our team had only one or two people for each function. Even in this situation, we were still determined to follow up on the development of chips using the 16nm process, and from the end of 2014 to the beginning of 2015, we completed the development of the first generation of 16nm chips with the internal code name BM1386. Of course, due to many reasons such as the market environment, this chip was not put into production in the end, but the accumulation of these technologies actually laid a very solid foundation for the success of BM1387.

"put all one's eggs in one basket"

The development of 16nm chips was put back on the agenda in the second half of 2015, when we started developing a project with the internal code name BM1387.

From a macro perspective, this time point is actually a recovery phase after the market hit the bottom in early 2015. All aspects have improved, but the crisis has not been completely eliminated. Therefore, the brothers and sisters in the chip team are well aware of the heavy burden on their shoulders. At that time, the team was still a very small scale that could be counted on two hands. Everyone had a heavy task, and there was no rest during the several holidays in the second half of 2015 - Mid-Autumn Festival, National Day, and New Year's Day. But everyone still had no complaints, and worked overtime almost every day until late at night, and even worked all night for several consecutive days in the final stage. It is such a highly responsible and enterprising spirit that supports the progress of the entire project. I guess everyone may also feel that we actually have no way out.

I still remember the night when I uploaded the layout file to the TSMC system. We drank and chatted until dawn. After dawn, no one left in a hurry, but no one was still sober.

"Good things come with a lot of hard work"

My memories after that are mostly small episodes. After the Lunar New Year in early 2016, we were waiting for the production of chips. During this period, an earthquake occurred in Taiwan, and we were worried that the chips would be affected. The good news is that our first batch of chips produced during the earthquake successfully completed the functional test. The bad news is that the performance was slightly affected. It was not until we received the second batch of chips that we officially unveiled the first generation of king chips.

Although the indicators of this chip seemed to be far ahead of many existing products on the market at that time, the technical team did not stop there. Before the mining machine was actually mass-produced, it also optimized the process and packaging, so that the indicators of the mass-produced chip were taken to a higher level. This is the appearance of the chips on most S9 mining machines on the market today.

To sum up, I always feel that no success is accidental. Even if these inevitable factors are not so clear in the process, they still exist. This may be the essence of the so-called success, and it is also the birth of the BM1387 chip.

02

The entire research and development process of S9 was quite tortuous, because S9 used a brand-new control board, and the 1387 chip also used brand-new interactive instructions at the time. The entire software was equivalent to being completely rewritten, and almost all the pitfalls were encountered. The most impressive thing was that the configuration would be lost after restarting. At that time, almost all the relevant codes of the file system were read through, and finally the problem was solved with less than 10 lines of code. Major problems are often caused by minor mistakes.

At that time, all S9 R&D colleagues were in Shenzhen, living a 007 life, working more than 16 hours a day. They worked hard for more than two months, and many people's hair was scratched out, haha.

S9 is the first mining machine that uses the sweep frequency mode. During the development of the sweep frequency mode, I got married. My boss called me at the wedding to urge me to make progress. I took a 3-day wedding leave and then rushed to Shenzhen. I didn’t have a honeymoon or travel, and I have been complained about it ever since...

S9 was a brilliant start, and then it was upgraded to S9i, and later to S9j. The parameter changes from S9i to S9j here are not big on the surface. For example, the change of J/T is only from 94 to 93. However, this change of 1J/T can be called a "blood and tears history". In order to strive for a little upgrade, software engineers even moved the mining machine up and down, unscrewed the screws, and had all kinds of skills. After debugging, their arms became thicker. For the fan adjustment alone, in order to achieve better performance at various temperatures, engineers put more than a dozen machines in the incubator, squatted beside them to repeatedly adjust the temperature of the incubator, and observed the performance of the fans. They squatted for fourteen or fifteen hours, and had to open the incubator from time to time to debug. It was either a cold current of dozens of degrees below zero or a heat wave of 40 or 50 degrees. Everyone jokingly called this task "refining the elixir". It takes two days and two nights to make a furnace of elixir, and I am afraid that dozens of furnaces of elixir were refined. After several hundred prototypes were debugged and tested repeatedly for months, we finally achieved this 1J/T improvement. Just think about the electricity bills and lunch boxes that can be saved for everyone, it is worth it~:D

03

The supply chain is a large line that includes quality, engineering, procurement, planning, after-sales, logistics and other aspects, and plays a vital role in the production process of mining machines.

In 2017, Bitcoin ushered in a bull market. The level of customer demand for products was constantly being refreshed. The growth rate of production capacity demand was unprecedented in the entire manufacturing industry. In the third and fourth quarters, it even grew exponentially, with the peak reaching one million units per month! This is more than 10 times the monthly production capacity predicted in our OKR target plan at the beginning of the year, which was unexpected by everyone. However, in the face of arduous difficulties, the entire supply chain was united, customer-centric, and returned to the original intention of hard work, completing the company's order tasks in a timely manner and meeting customer expectations. Many stories were born in this process, all of which are impressive——

"Planning and Production"

Since Q3 2017, we have been increasing production capacity at a rate of doubling every month. This stage can be said to be the most painful. No weekends and working overtime until 1 a.m. every day are the norm. At that time, the new factory was only one and a half months from rough decoration to production. The corresponding living facilities of the factory had not yet caught up. The meals of the employees were prepared in the kitchen 1 km away, and then pulled to the inner yard of the factory by truck. The tailboard was used as a distribution table. Everyone brought their own lunch boxes and ate standing on the spot. This lasted for more than two months. The fastest growth in production capacity was in October. The production capacity plan was to double from the original 5K/day every week. As a result, we achieved a daily output of 40,000 in 3 weeks! To achieve this goal, all our technical and management cadres went to the front line to support it. We made great efforts to achieve the results by ensuring that abnormalities of production equipment were solved immediately on the spot and new employees were trained in the "four-step method" as soon as possible.

When production capacity exploded, all resource coordination had to be done in a planned and step-by-step manner. Any slight mistake could immediately lead to a complete stop in delivery. So we adjusted many coordination management methods, such as increasing the daily delivery frequency of outsourcing by 2-4 times; in order to ensure that all loading and unloading docks in the factory could smoothly complete the entry and exit of raw materials and products, we arranged special personnel to dispatch the order of entry and exit, and worked day and night shifts. During the peak period, the factory had more than 100 vehicles entering and leaving in a single day, and the queue of vehicles exceeded 2KM. The prosperous scene of heavy traffic at that time is still vivid in my mind.

04