“Overachievement”: Xiaomi's Decade-Long Chip Development Journey

Today’s post is a translation and transcreation of an article published by Jiemian News, a Chinese language tech outlet, that profiled Xiaomi’s homegrown chip-making backstory. While I rarely do translations, this article was sufficiently interesting and included some fascinating insights and anecdotes behind the overnight success of Xring O1, Xiaomi’s recently released flagship chip, that was a decade in the making. (For a more technical deep dive on the chip itself, I recommend this post by Nomad Semi)

Along this post, I added a few parentheticals and annotations to contextualize certain parts of the story for readers, who aren’t familiar with either Xiaomi the company or China’s semiconductor industry boom and bust in the last five years. As you will soon see, Xiaomi was a victim initially of China’s nationwide pursuit for semiconductor self-sufficiency because it misdirected capital and talent into many failed startups and corrupt outfits. Ultimately, Xiaomi’s execution and conviction, backed by sound business logic (not necessarily government mandate) won out.

This post was 70% Claude Opus 4, 30% me. Hope you enjoy it!

---

On May 23, 2024, at 8:03 in the morning, a call that turned out to be worth tens of billions rang on Lei Jun's cell. (Lei Jun is the founder, chairman, and CEO of Xiaomi Group.)

On the other end of the line, in a lab at Xiaomi's Shanghai R&D Center, over a hundred people stood crowded together, holding their breath, waiting to witness their own historic moment. On the previous day’s afternoon, this core team of what’s internally known as the New Business Department at Xiaomi, had just received the first batch of 3nm chips back from tape-out. The whole team pulled an all-nighter to finish preliminary verification of these chips on smartphones in less than 24 hours. This step showed the successful tape-out of the "Xring O1" chip.

After the call was picked up, the person speaking was Zhu Dan, Vice President of Xiaomi Group and head of the "New Business Department." He reported this good news to Lei Jun. While congratulating them, Lei Jun told the team this milestone was "quite remarkable." The call was then made to Lin Bin (a co-founder and vice chairman of Xiaomi), Lu Weibing (president of Xiaomi), Zeng Xuezhong (senior vice president and president of the Xiaomi’s International Business Department), and others.

On this day, a huge weight finally lifted from the shoulders of Xiaomi's semiconductor R&D team.

That call was an important milestone. It proved Xiaomi's success in developing its own 3nm mobile SoC chip, making it just the fourth mobile smartphone maker in the world with this capability, after Apple, Samsung, and Huawei.

One year later, on the evening of May 22, Lei Jun stood on the main stage of the Beijing National Convention Center to officially tell this decade-long chip-making story: In 2014, Xiaomi Pinecone was established (its chip R&D subsidiary); in 2017, the Surge S1 was released (its first homegrown chip); in 2019, its chip development progress hit a wall and pivoted to focus on "small chips"; in 2021, the company reignited its “big chip” initiative; in 2025, Xring O1 was launched.

In the buildup over the previous days, public opinion had already divided into two camps: one was excited about Xiaomi's resurgence in cutting-edge chip self-development; the other constantly raised questions about the value of building its own chips.

Lei Jun didn't hide the fact that Xiaomi's chip-making efforts were still lagging, while striving to squeeze into the first tier. But there's no doubt that this is a necessary path in order for Xiaomi to fight for a spot in the high-end flagship smartphone market. It also reflects Xiaomi's determination to reach its ultimate dream of being a truly “hardcore” tech company.

"Don't expect us to crush it and dominate right off the bat — that's impossible," Lei Jun said. "But if you see areas where we surpass Apple, please give us a shoutout, because even a little bit of surpassing is very difficult."

The Restart

In early 2021, Xiaomi's management made two strategic decisions that would determine the company's direction for the next decade. One was building electric vehicles, which was soon made public, and the other was making chips, which at the time had been dormant for four and a half years.

Xiaomi had already stumbled once in chip-making. In 2014, Xiaomi established Pinecone with the same determination of long-term investment and began developing mobile SoC chips. But ultimately, the 28nm Surge S1 wasn't validated by the market, Surge S2's release was indefinitely postponed, and Xiaomi's chip-making effort was paused.

However, during the intervening years when that effort appeared totally abandoned, Xiaomi didn't stop its involvement in chip R&D. At the end of 2017, Xiaomi Ventures was established. According to Tianyancha data, by April 2021, it had directly invested in 45 semiconductor companies.

Until 2021, Xiaomi decided to restart its “big chip” development effort. However, there was hesitation and confusion within Xiaomi about which specific device to tackle first. There were many options at the time: Should they go for a mobile flagship smartphone SoC, a flagship tablet SoC, or automotive chips like smart cabin or smart driving chips? After many rounds of discussion, they discovered that what both Lei Jun himself and the core chip business team really wanted to pursue was the mobile flagship smartphone SoC — the crown jewel of chip-making — even though this path would also be the most difficult.

Xiaomi's situation at the time was completely different when it decided to rekindle its chip-making ambition. In 2021, Xiaomi had just chosen to go after the high-end market the previous year and achieved a 20% year-over-year sales increase, becoming the only smartphone maker among the top five besides Apple to have growth, confidently becoming a top-3 smartphone maker globally. To continue moving up, mastering chip-making became a necessity.

Examining the failure of Pinecone, Xiaomi learned several lessons. It was a joint venture formed outside of Xiaomi, and with the company located in Nanjing far away from the headquarters (Xiaomi’s HQ is in Beijing), communication and collaboration between teams were not smooth enough, making it difficult to fully integrate systems and chips.

"For a phone company making chips, honestly, this is a major taboo," said a Xiaomi insider. "There must be no barriers between departments, and interests and ideas must be completely aligned. It must exist as ‘One Team’.”

Therefore, when Xiaomi returned to chip-making, it chose to build a first-level department called the New Business Department under the mobile phone division, keeping it "inside" rather than "outside" from day one. From the start, it was led directly by Xiaomi Group Vice President, Zhu Dan. As a former Motorola R&D engineer, he joined Xiaomi in 2010 as employee number 54 and had been responsible for several important business units over the years.

At that time, Zhu Dan's primary task was to build the team. The initial team mostly came from internal transfers, while recruiting talent externally wasn't easy. Xiaomi's chip-making restart was kept a secret, which created a lot of communications friction with candidates. An R&D manager once lamented that while Xiaomi Group had officially announced its EV making plan, it hadn't publicly shared the restart of its chip making plan, so when he called many people to recruit them, no one would believe him.

Moreover, 2021 happened to be a boom time for domestic chip startups, when many seasoned semiconductor pros who left large companies opted to join new startups rather than Xiaomi, which had failed once before. By the end of 2021, Xiaomi's New Business Department had only two to three hundred people.

But the environment suddenly shifted in 2022, when early stage investors began turning a cold shoulder towards semiconductor startups. That year became a critical time for Xiaomi to recruit talent because many startups could no longer sustain themselves.

"The semiconductor circle is small. Many people began to find out that certain companies might not be that legit. After another year of observation, they found that Xiaomi was really serious about making chips, and if you look at the company’s foundation and business logic, Xiaomi indeed has this capability," the aforementioned insider told Jiemian News.

That year, many semiconductor veterans joined Xiaomi, and by year-end, the team size was approaching a thousand. According to Jiemian News, among the current 2,500-plus people in Xiaomi's chip team, most are R&D personnel. The number of team members who came from either internal transfers, external recruitment, or university recruitment each account for about one-third.

(Annotation: from 2020 to 2022, China went through a rapid boom and quite a bit of bust in its domestic semiconductor industry, choked full of government subsidies, talent poaching and rampant corruption. I wrote about one of the most egregious cases three years ago, the Wuhan Hongxin Semiconductor Manufacturing, which scammed close to $2 billion dollars and even lured the founding CTO of TSMC to join it temporarily. Little did I or anyone else know, Xiaomi was also trying to rekindle its chip-making ambition at the time, and eventually benefited from some of these busts and put talent to good use.)

Homegrown R&D

After Xiaomi officially restarted its “big chip” R&D initiative, it set three goals: high-end flagship processor, most advanced process technology (in terms of nanometers), and first-tier performance.

A person from Xiaomi's New Business Department believes that at the core of Xiaomi's chip-making restart lies a key business insight. "For example, to focus directly on a flagship SoC is a very important insight." He believes that phone companies that make their own chips must start with the flagship SoC, not mid-range chips. The market condition for the mid-range category is sufficiently competitive, thus phone companies don't need to make another chip for it because there is no room for differentiation.

At that time, the entire Xiaomi executive team was fully prepared to fight this battle. According to a senior executive in Xiaomi's chip team, when he first met Lei Jun before joining, Lei Jun brought a tablet computer. During their nearly two-hour conversation, Lei Jun's questions were very deep and detailed, and he took notes while talking. His biggest takeaway at the time was that Xiaomi was serious about making its own chips and had accumulated sufficient insight to do so.

However, insight alone isn't enough. After the rubber meets the road, Xiaomi needs to answer one question: How much value is there really to make its own Xring O1 chip?

Looking at the chip design, Xring O1's CPU is a 10-core 4-cluster architecture, all using Arm's public IP. Using public IP is also the most mentioned critique among Xiaomi skeptics. IP here refers to the modules with specific functions inside the chip. On SoC chips, assembling IP is like placing Lego blocks. How to build a layout with better energy efficiency advantages in limited space with the same blocks is key to chip design.

The fact of the matter is, using public IP is a path that almost all companies with internal chip R&D capabilities have taken initially, including Apple, Qualcomm, and Samsung. Xiaomi insiders told Jiemian News reporters that increasing the proportion of self-developed IP is certainly a long-term pursuit, but no one can achieve it overnight. For example, Apple's first homegrown chip, the A4, received company-wide investment, but was ultimately still highly reliant on Samsung’s technology. Apple only began using its homegrown CPU with A6 and homegrown GPU with A11, gradually becoming a tier-1 chip-maker with "Arm proprietary IP + homegrown core” and eventually achieving more superior performance.

In terms of design work of its first-generation flagship SoC, Xiaomi actually achieved a lot of breakthroughs. Even public IP is not something that is out of the box ready to use, and anyone can use it well. It requires a lot of backend design innovation and iterative layout optimization, which is a massive amount of work in order to achieve a good performance and energy consumption profile.

For example, with Xring O1, more than 480 of the standard libraries of the CPU module were modified or redesigned. This quantity almost reaches 1/3 of the total number of 3nm process standard libraries. These libraries are the smallest logical units inside a chip, akin to the individual bricks that build skyscrapers. This level of engineering and redesign cannot be done without solid technical foundation and knowledge.

Additionally, Xring O1's ultra-large core has a maximum frequency of 3.9GHz, reaching the highest commercial frequency of X925. Jiemian News reporters learned that Xiaomi's chip team did a lot of backend optimization and low-level tuning and iterated over hundreds of versions to finally achieve this high frequency goal.

Counterpoint senior analyst Ivan Lam told Jiemian News that when evaluating a smartphone company’s homegrown chip R&D progress, Xring O1's current approach of using self-developed application processors combined with external third-party baseband chips is sensible and understandable. Globally, except for Huawei and Samsung, no other smartphone manufacturer has made breakthroughs on baseband integration capabilities.

According to published teardown tests, Xring O1's CPU load energy efficiency performance is better than the Dimensity 9400 chip (from Mediatek), which also uses Arm’s public IP, and close to that of the Snapdragon 8 Elite (from Qualcomm). Its GPU energy efficiency performance, however, is still inferior to both. But its overall performance can be considered first tier.

In summary, Ivan Lam believes that as a first-generation product, Xring O1’s main mission is to gain market validation on a technical level.

Mounting Pressure

Eight years ago, when the Surge S1 was released, Lei Jun once said, before going into tape-out he repeatedly asked the team, "Have you really tested it? Is there really no problem?" Tape-out is a key step in verifying whether a chip design meets expectations. One tape-out typically costs tens of millions of dollars and takes more than three months.

With Xring O1, Lei Jun didn't put that much pressure on the team. This foundation of trust was built bit by bit. At the beginning of the project, Zhu Dan led the team to meet with Lei Jun on a weekly basis. During the most intensive times, the meetings happen three or four times a week.

In January 2024, Xring O1 officially went to tape-out. In May, the batch came back. Over the following year, the New Business Department went through a long verification and tuning process, including several months of testing the chip's stability, software hardware adaptation development, and more. The final Xring O1 achieved a benchmark score of 3 million in the lab. The Xiaomi 15S Pro phone, which will debut with Xring O1 inside, achieved a single-core performance score of 3008 points in real-world room temperature testing, compared to Apple A18 Pro's 3529 points. The multicore testing score exceeded 9500 points, slightly better than the A18 Pro.

However, although performance exceeded many people's expectations, Xiaomi's chip team still faces enormous pressure at this moment. They will closely monitor user feedback after the product launches, which will involve many scenarios like heat, power consumption, and many other aspects. Only if the product sells well can the future of the homegrown chip project be supported.

At the launch event, Lei Jun did the math: for a flagship chip like Xring O1, if only 1 million units are sold, the average chip R&D cost per phone would exceed $1,000 dollars. "Without sufficient volume, even the best chip is a money-losing business."

This kind of pressure has actually been with Xiaomi's chip team all along. As an organization that grew rapidly in the last four years, Xiaomi's New Business Department faced an all-encompassing challenge of having to build the team and establish the processes, while also defining the product and developing the technology at the same time. From multiple angles, it is hard to find a fitting historical precedent in the semiconductor industry.

But during the four and a half years of R&D, Lei Jun never issued any military-like orders (军令状). He instead occasionally tried to relieve some pressure from the team. "Making chips requires steady progress, don't expect overnight success. If you can make a first-generation product without ruining our brand or reputation, that would already be considered a success."

Going back two years to May 12, 2023, Zhu Dan was meeting with key suppliers that day when suddenly news broke that OPPO's Zeku team (OPPO’s homegrown chip division) is shutting down and the company is terminating its chip making effort. This chip design subsidiary, established less than 4 years ago with more than 10 billion RMB invested, disbanded its 3,000-person team on the spot.

Although it's normal for chip making ventures to have a high failure rate, Zeku's sudden closure inevitably impacted the morale of Xiaomi's chip team. Lei Jun quickly called Zhu Dan to convey a simple message to everyone on the team: we are committed to continue making chips.

A few days later, the New Business Department held an all hands meeting to internally communicate the business logic of "taking ownership" of chip-making, and why external industry changes would not shake their conviction. An employee recalled somewhat jokingly that when everyone went to the meeting that day, they urged each other to quickly swipe their access cards to see if they still worked.

At Xiaomi, chip-making is a long-term, "50 billion investment over 10 years" kind of plan. So far, the investment has already exceeded 13.5 billion RMB with this year's budget allocating another 6 billion RMB. Lei Jun stated that Xiaomi will resolutely shoulder the pressure, and in the next five years, invest another 200 billion RMB in core technology R&D.

Due to confidentiality restrictions, many of Xiaomi's chip-related patents and intellectual property have only recently entered the application stage. Although many people have had SoC development experience, in their view, the difficulty and excitement of building Xring O1 still stacks up as a top-3 career experience.

"Internally, we all discussed repeatedly, including colleagues who came from various companies, and everyone unanimously agreed that for each and everyone of us, we overachieved with Xring O1. It was a rare experience of going from 0 to 1, under a complex industry environment and technical conditions."

On the evening of the May 22 launch event, a tech blogger's video called "In-depth Review of Xiaomi's Homegrown Xring O1 Chip" was livestreaming on Bilibili. This was the first video on the entire Internet that was doing a teardown analysis of Xring O1. Traffic surged instantly with peak live viewership of 100,000 people, fully reflecting the outside world's curiosity about the real performance of Xiaomi's homegrown chip.

Xiaomi's chip team can breathe a brief sigh of relief, but reality won't allow them to relax for too long. Some employees will soon go to Xiaomi Home Stores to personally sell the Xiaomi 15S Pro equipped with the Xring O1 chip to consumers — that's where they will truly face the market's test.