Tech war: how Chinese scientists rigged a low-cost AI computer chip to power a hypersonic weapon
- Using a Nvidia module widely available online, Chinese researchers overcome significant obstacles to boost performance of hypersonic vehicle
- Jetson TX2i chip installed in the scramjet engine control system boosted range, stability and significantly cut research and development costs
Tests have suggested that this particular module can process computational fluid dynamics models with unprecedented efficiency, meaning calculations that previously took seconds to complete could now be done in just 25 milliseconds – four times faster than the blink of an eye.
The module’s response speed made it ideal for “real-time optimisation of the fuel supply system, fault diagnosis, and fault-tolerant control in scramjet engines”, according to a joint project team from Beijing Power Machinery Research Institute and Dalian University of Technology. Their peer-reviewed paper was published on March 13 in the Chinese academic journal Propulsion Technology.
US ‘may be sending strong message’ to China with hypersonic missile test
Headquartered in the United States, Nvidia is the world’s largest supplier of AI chips. Nvidia began selling the TX2i for industrial applications about six years ago. The module’s peak single-precision performance is 1.26 TFlops, about one-fiftieth of the capability of the company’s most powerful AI chip, the H100.
The H100, however, costs tens of thousands of US dollars and is in short supply. The TX2i, on the other hand, can be obtained for a few hundred dollars, is not subject to US export controls and is widely available online.
When contacted by the South China Morning Post on April 12, Nvidia said it had no comment on the matter.
The project team, led by Professor Sun Ximing, said in their paper that the TX2i module in the scramjet engine control system not only boosted the range and stability of hypersonic vehicles, but also significantly reduced their research and development costs.
“High-performance graphics cards possess excellent computational capabilities but require supporting equipment such as a hosting platform, power supply and radiator.
“They have disadvantages such as high power consumption, heavy weight and large size, which do not meet the demands of lightweight and small-sized embedded controllers in the aerospace field,” Sun’s team wrote in their paper.
Because of the sequential nature of hypersonic flow field simulations, where one event must occur before another can be calculated, industry experts generally believe that such computational tasks cannot be accomplished using lower-end AI chips adept at performing simple parallel computations.
To solve this problem, Sun’s team introduced a novel CPU plus GPU architecture, which detailed how to tackle sequential parallel computing problems by ensuring the two different chip types worked well together.
The “step-by-step” guide provided in the paper provides detailed formulas and addresses potential engineering challenges, including limiting simulation grid size, memory management, code optimisation, and specific compilation instruction schemes.
02:44
Iran’s new missile can reportedly reach Israel and US military bases in Middle East
For better universality, the engine controller’s interface and communication protocol adhere to international standards.
However, the engine is only one component of the entire weapon platform.
“To apply [the AI chip] to hypersonic vehicles, further work is needed in inlet modelling, shock wave correction, and data modelling,” the team said in the paper.
Some important parameters involved in these tasks usually need to be obtained in extensive wind tunnel testing and actual flights.
It remains unclear why the researchers selected the Nvidia chip for their experiment, and the authors could not be reached for comment.
It is possible that the intention was to prove the feasibility of using an inexpensive AI chip for hypersonic weapons, regardless where it is made. However, such weapons can operate with different types of chips.
While few countries are able to design and manufacture such chips, a growing number of nations, including Germany, France, Japan, North Korea and Iran, have launched hypersonic weapon research and development programmes.
The proliferation of hypersonic weapon technology has been a major concern for the United States. In 2017, the Rand Corporation suggested that Washington work with Moscow and Beijing to prevent other countries from acquiring such technology.
But while Chinese and Russian military experts admit hypersonic weapons technology poses certain risks, they said it is more likely to accelerate the collapse of the US-centered world order.
They have reasoned that hypersonic missiles could penetrate the defences of America’s massive aircraft carrier fleet, which it has long relied on for global military superiority.
In a war game exercise conducted last year by Chinese scientists, a Ford-class aircraft carrier battle group was destroyed by about 20 hypersonic anti-ship missiles. If more countries possessed hypersonic weapons, the advantage enjoyed by the few countries with significant naval power over the past five centuries may come to an end, according to some experts.
North Korea says it test-fired new solid-fuel hypersonic missile
On April 3, North Korea successfully test-fired a hypersonic gliding missile, and on Monday, Iran claimed to have used hypersonic missiles to attack Israeli military bases in retaliation for Israel’s bombing of the Iranian embassy. Currently, there is no evidence to suggest that China or Russia provided weapons or technical support to either country.
