What’s in a millisecond? The difference between today’s embedded systems and tomorrow’s according to Professor Gerhard Fettweis of the Technical University of Dresden in a speech that he gave at the very beginning of the DATE conference this morning of which is he is the general chair.
Why 1ms? Fettweis argued the driver is what he called the “tactile internet”, contrasting the need for low-latency responses versus what is possible over current networks. Users moving a mouse or using touch to drag a pointer around a display expect the system to react within 1ms, he said. But the best-case latency over a network such as LTE is 25ms.
“We have to get systems to go down to 1ms to get to these tactile interfaces,” Fettweis said. “Gamers know this already. We also know this from robotics. They have to be controlled with a 1ms round-trip latency. The reason why 1ms is so important is because the systems have to suppress the dominant resonant frequency by operating the control loops at least ten times faster. For a 3m-long object, this is in the range of 100Hz, which leads to a round-trip latency of 1ms,” Fettweis explained.
Because of the trend to mass customisation or individualised production, Fettweis argued highly mobile, connected robots will become much more important. “Conventional assembly lines don’t work. We will have robots running around on the factory floor and assembling things. They will be like hummingbirds, moving between different assembly stations.
“In this environment, countries with a high level of personal compensation will not be at a disadvantage because the production systems will all be robotic. But we need those 1ms tactile latencies.”
Low-latency needs expand
As robots are used to operate in hazardous environments, such as the Fukushima cleanup, they will also need similar low-latency responsiveness to allow easy remote control, Fettweis said.
Fettweis argued the same need for low latency extends to cars, especially as they use wireless communications to form platoons as they drive along highways. “The platoons only work if the cars talk to each other. The speed of interaction has to be 1ms because distances are on the order of 3m,” he argued.
As a result, the emphasis on delivering content in the 4G wireless standard will shift to supporting steering and control in the coming 5G systems, Fettweis claimed. “Many things in industry, healthcare and automotive will be reinvented when the tactile internet happens. The good news is that they are strongholds of Europe. But we need massive computing to support it.”
Fettweis pointed to TU Dresden’s work on the HAEC box as an example of how computing design will change through the use of stacked chips in a 3D architecture and short-range wireless communications. “This is the vision we are heading for.”
As the next decade gets underway, Fettweis said he expects each processor chip to carry more than 100,000 individual processor cores. “And we will have 128 chips stacked on top of each other – that’s eight times more than in the current iPhone. A single board designed to go into one litre box would carry a 4 x 4 array of these stacks, with four boards able to squeeze into the cube. The result, he said, would be a one billion-processor system, delivering 100,000 more performance that today’s machines.
“It means we still have 25 years of advancement at the speed of Moore’s Law using these technologies. And this is a gigantic chance for Europe. Europe owns 30 per cent of the embedded systems market. The area consumed by silicon will go up by more than a factor of ten. If you have 16,000 chips in the space of an internet access point, that is at least two orders of magnitude more than today,” Fettweis said.
Referring to the blueprint for European semiconductors to double local chip production that the European Commission is trying to develop at the behest of commissioner Neelie Kroes, Fettweis said: “Neelie Kroes’ ambition for Europe to gain 20 per cent of world IC production is not aggressive enough for what I am talking about here.”