Shenzhen, China

How China is aiming for a state of technological independence

Image credit: Image Source

Rather than hampering semiconductor innovation in China, is America’s technology policy actually boosting it?

China’s long-standing bid for technological independence, including its semiconductor ambitions, are again climbing the global agenda.

In the West, the Trump administration’s clampdown in August on Huawei’s access to US technology highlights – and reflects – concerns about China’s goals as stated in programmes like Made in China 2025.

Speaking in October 2018 at the Hudson Institute, Vice President Mike Pence signalled just how strongly the US now views tech competition with the Middle Kingdom.

“To win the commanding heights of the 21st century economy, Beijing has directed its bureaucrats and businesses to obtain American intellectual property – the foundation of our economic leadership – by any means necessary,” he said.

“Beijing now requires many American businesses to hand over their trade secrets as the cost of doing business in China. It also coordinates and sponsors the acquisition of American firms to gain ownership of their creations. Worst of all, Chinese security agencies have masterminded the wholesale theft of American technology – including cutting-edge military blueprints.”

Meanwhile in China, there remains great anger, though not that much shock, over Huawei’s formal addition to the US Entity List.

Its industry had earlier seen ZTE, Huawei’s local telecoms infrastructure competitor, nearly driven into bankruptcy by US-imposed sanctions in April 2018. Although a deal was struck, Pence and President Donald Trump continued to raise the pressure on Beijing. It was clear there was and is more to come.

Moreover, China’s fears over technological subservience to the US long preceded that.

Just a year into his presidency, Xi Jinping addressed the June 2014 joint meeting of the Chinese Academies of Science and of Engineering. “Only by mastering crucial core technologies with our own hands can we fundamentally safeguard our national economic security, national security and security in other areas,” he said.

Xi and other senior state figures have issued the same warning in many speeches since then.

Surveying this landscape, many leading Chinese technology companies have begun to work on the assumption that what happened to Huawei could soon happen to them.

And developments in the last few weeks may be encouraging others to follow suit, supporting the view that Washington is ready to spread the net.

For the first time, a senior US State Department official has publicly criticised China’s powerhouse ‘BAT’ (Baidu, Alibaba and Tencent) triumvirate of internet companies and their business practices. In September, a cross-party Congressional group tasked the Pentagon with creating a list of companies operating in the US that have links to the Chinese military and which present an “economic espionage” risk.

The Sino-US standoff raises many questions, including, from an engineering perspective, those around China’s ability to fill any technology gap that may widen between it and the US. Here, semiconductors are regarded as one of the country’s weak points.

“Despite 40 years of effort, investment, and espionage, China is unable to make advanced semiconductors,” writes James Lewis, in a report for the Center for Strategic and International Studies (CSIS) on China’s drive towards silicon self-sufficiency.

“There have been embarrassing frauds... and expensive failures, such as the centrally directed programs of the early 2000s that wasted billions of dollars in building unproductive semiconductor fabrication facilities.”

A caveat must apply to Lewis’s observation: China does not have a complete domestic supply chain for the most advanced chips. So, Huawei can design them (shown by the AI capabilities of the Kirin family developed at its Hisilicon subsidiary), but they must go to Taiwan’s TSMC, the world’s biggest chipmaker, for manufacture.

Nevertheless, Huawei has so far only been able to place such silicon within wider systems by working with partners that have close US links.

These include component suppliers such as Qualcomm, processor IP suppliers such as Arm (though UK-based and Japan-owned, Arm’s US development activities are subject to Entity List restrictions), and EDA companies that supply design tools.

Huawei’s dependency on those parts of a global supply chain, and the consequences of its disruption, imply that the company has been pushed back towards Lewis’s blanket definition of China’s semiconductor capabilities.

However, Huawei may not be entirely out of alternatives and nor may China as a whole.

From a design perspective, three options stand out: forking existing technology, switching to available non-US-controlled alternatives, or building new technology from the ground up.

The first two are the preferred short-term responses. The third is seen as a longer-term solution where there is promise in emerging/leapfrog technologies for, say, quantum or neuromorphic designs.

Huawei has hinted at the possibility that it could fork Arm’s IP.

The company says that it has a “perpetual” and “architectural” licence to the current v8. Arm is constrained from collaboration with Huawei and from providing it with updates or further versions of its IP as long as the full ban is in force. However, Huawei’s current position is that the licence remains valid for existing chip designs (including some not yet on the market).

This had led analysts to wonder if the company might go further and assert that the terms mean it can also continue to enhance and augment the v8, if the work is done internally. Huawei and Arm have previously had a long relationship, so Huawei’s internal understanding of Arm’s technology is considerable.

We are not at that point; the US and China could sufficiently resolve their differences to get some, maybe all, of the Entity List restrictions lifted.

In the meantime, though, other major Chinese players are exploring alternative technologies. The main semiconductor option here is RISC-V open-source processor IP. Although originally developed in the US at University of California - Berkeley, everything a user needs to work with it is available in the wild and can no longer be tamed by Washington.

‘Only by mastering crucial core technologies with our own hands can we fundamentally safeguard our national economic security, national security and security in other areas.’

President Xi Jinping, China

A technology that was once seen as struggling to gain traction in China now has 25 companies participating in the RISC-V Foundation that oversees the technology, and nearly 200 within domestic research and advocacy groups.

Some of the results have been impressive. In August, Pingtouge, Alibaba’s silicon subsidiary, unveiled what it claims is the most powerful RISC-V implementation so far according to the CoreMark benchmark. The XuanTie 910 is a 16-core 64-bit RISC-V processor that targets system-on-chip designs for 5G, AI and Internet of Things.

It followed this in September with the Hanguang 800, an entirely in-house-developed AI inference chip that it claims is 15 times more powerful than Nvidia’s T4 GPU, and which initially at least will be used on Alibaba’s own cloud services.

Pingtouge was launched by Alibaba’s then executive chairman Jack Ma in September 2018, before Washington’s moves but after those against ZTE, which were seen as partly inspiring the move.

Pingtouge did integrate RISC-V expertise. Alibaba had previously acquired C-Sky Microsystems, a Hangzhou design house that had been working with the Instruction Set Architecture (ISA). The speed with which Pingtouge has delivered the XT910 is noteworthy.

The implementation adds 50 non-standard instructions to RISC-V’s base set of 32 for tasks such as managing memory and wrangling the CPU cores. It also has more efficient out-of-order software execution; typical RISC-V implementations have in-order execution.

There are then two other significant aspects to the XT910. At the launch, Pingtouge said that the design targets the 12nm fabrication node and will be open-sourced (though full terms are still awaited).

The 12nm node will soon be available through China’s largest domestic fab, SMIC. It says it has completed development of the process and is now in “customer verification”. Open-sourcing could therefore allow other Chinese companies to work towards complex use-cases with the XT910 inside a domestic supply chain.

Pingtouge is not the only Chinese company offering RISC-V rivals to ‘traditional’ platforms.

In September, Beijing-based GIGA Device announced a lower-end microcontroller also based on the ISA. It is based around a ‘Bumblebee’ RISC-V core the company has developed with another local company, Nuclei System Technology. Importantly, it is marketed as a direct 32-bit alternative to GIGA’s Arm-based GD32 MCUs – with claims of 15 per cent better performance at half the power.

GIGA Device says it will remain an Arm licensee but wants to give customers a choice.

And on it goes. Chinese consumer electronics brand Xiaomi offers a healthcare watch based on a RISC-V AI chip from local partner Huami, the Huangshan No 1. SiFive, holder of the RISC-V crown before Pingtouge, has separated its Chinese R&D and commercial operations from those elsewhere as it continues to work with the technology. It is not hard to see why.

All this suggests gathering momentum behind RISC-V based around China’s drive for independence in hardware - and yet as quickly as China is innovating around technologies such as RISC-V, many analysts continue to see the government’s continuing focus on manufacturing as a possible brake on innovation. SMIC is, after all, still to formally offer the 12nm node; Taiwan’s TSMC, the world’s largest foundry, launched it in 2016.

TSMC reacted to Huawei’s addition to the Entity List by saying that it could still do business with the company and remain in compliance with the US restrictions. That is good news for other cutting-edge players on the mainland. For now.

A question mark remains over Beijing’s semiconductor spending. The second iteration of its National IC Investment Fund still makes fabs a priority, though it has a greater design component than the first.

Total planned investment – including that from local government – is $118bn (£94.4bn) over five years. CSIS’s Lewis notes that this needs to be seen against the $13bn (£10.4bn) that Intel alone invests annually to get some idea of the challenges China faces in catching up in arguably the most complex of all manufacturing sectors.

Remember the scope of Beijing’s ambitions here. China made only 16 per cent of all the chips used there in 2018. The government has a 2025 target of 70 per cent.

It is an issue. But it doesn’t necessarily mean that China is not moving in the right direction – just perhaps not as fast as its leaders would like in every respect.

Robert Atkinson and Caleb Foote of the Information Technology and Innovation Foundation recently analysed decade-long comparative performance across 36 innovation benchmarks for China and the US. They identified progress on China’s side in every one, albeit to varying degrees.

“If China is only a copier, then the risk to advanced economies is limited. But if China is more like the ‘Asian tigers’ that rapidly evolved from copiers to innovators, the threat is serious,” they write.

“As those nations became more innovative, they took market share from leading companies in Europe and the United States. There is no reason to believe China will not follow the same path, only with significantly greater impacts because the Chinese economy is massive, Chinese policies are more aggressively mercantilist, and it is much more difficult to get China to compete fairly.”

Atkinson and Foote believe there is reason to see China on that path and highlight a process originally described by Korean academic Linsu Kim in ‘Imitation to Innovation’.

Kim identified an interim development stage among the Asian tigers where they moved from copying to then developing innovations primarily for domestic consumption before moving on to compete in earnest globally.

China’s progress with RISC-V here can arguably be seen as essentially the same. So can that of players in other fields.

There are its mobile ‘mega-apps’ like WeChat, Tencent’s Sino-ubiquitous mobile platform that allows everything from messaging to paying utility bills to hiring a pushbike to even ordering and paying in a restaurant so you only interact with the waiting staff when food arrives or plates are cleared.

Or there is its work around the Kubernetes open-source containerisation technology that supports cloud-computing infrastructure, an important focus for all three BATs.

Going back to Huawei, there is its role in creating the standard essential patents for 5G. That cat is out of the bag.

In this light, the risks Atkinson and Foote identify may be more real than is often acknowledged. Has ‘decoupling’ already begun? Have the Trump administration’s policies even accelerated some branches of innovation rather than blunting China’s ambitions?

Not so much ‘interesting’ as ‘uncertain’ times. And given how strongly Beijing and Washington appear to hold their views, times that Chinese technology companies and the rest of us should assume will be with us for a good while.

 

 

Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.

Recent articles