2017 Mustafa(Pbuh) Prize laureate compares marketing academic startups to ‘slingshot around a black hole to get momentum’
A 2017 Mustafa(Pbuh) Prize laureate delivered a speech at MSTF-COMSTECH webinar .
MSTF Media reports:
Amin Shokrollahi, CEO and founder of Kandou Bus Co., gave a speech on “Data transmission at the lowest level” and talked about his personal journey in a webinar organized jointly by the Mustafa
Science and Technology Foundation (MSTF) and Standing Committee on Scientific and Technological Cooperation (COMSTECH).
Shokrollahi, Full Professor of Computer Science and Mathematics who has won the Mustafa
Prize for “Raptor codes,” addressed power and speed challenges and opportunities in data transmission during the webinar.
The webinar was opened by Coordinator General COMSTECH and a 2021 Mustafa
Prize laureate, Muhammad Iqbal Choudhary.
Choudhary introduced Shokrollahi to the audience, elaborating on his scientific background.
Next, Shokrollahi started his speech with recounting the story of his personal journey briefly.
He then began his lecture on data transmission and why it is important.
“Chips in an electronic device communicate with one another via electric wires connecting them,” he said, adding “the question is how to best send information across these wires─ best in terms of feed and power.”
Stating that “data centers are all about connectivity,” he showed a graph of power consumption in data centers.
As Shokrollahi pointed out, by 2030, networks, consumer devices, production of ICT, and data centers themselves will use about 21% of the entire electricity consumption in the world.
He displayed another graph including the energy forecast for Information and Communication Technology (ICT), which, according to him, is “a massive consumption.”
“In reality, I think these numbers are bigger,” he commented.
As the 2017 Mustafa
Prize laureate pointed out, there are 3 big challenges now.
The first challenge is that “Moore’s Law is pretty much finished.” Moore’s Law, he said, suggests that “about every two years, you can pack twice as many transistors on a given area at the same cost.”
Regarding Moore’s Law, he said “More performance for less power/ price is not anymore,” adding “Advanced manufacturing will continue to become more and more expensive and the end user is going to pay the bill.”
Another challenge is Geopolitical issues, he noted, giving the example of how tensions in Taiwan can seriously impact the supply chain.
The third challenge is the power used for communication between devices, he said, explaining that “the power used for communicating data increases very rapidly with the amount of data.”
“Without new fundamental innovations, I/O will become a major power bottleneck,” Shokrollahi continued.
He said the first and second challenges are out of his expertise and control, “but I may be able to do something about I/O; sending and receiving information.”
Shokrollahi, talking about the impetus of founding his company Kandou, said: “It all started with a casual conversation with one of my postdocs on how to solve a communication problem which had nothing to do with the one we ended up providing a solution for: communication between chips.”
Full Professor of Computer Science and Mathematics at this point of his lecture referred to the beginning of his talk and explained how high speed data transmission can be viable.
“As I said, chips in an electronic device communicate with one another via electrical wires connecting them. In high speed communication, groups of two wires are used to send one bit,” he said.
According to him, “If you have many wires, pairs swing independently and consume power.”
The fundamental idea he and his team had was sending information not across two wires, but across more than two wires; across three or four wires.
If pooling more than two wires is done “correctly,” more bits per wire without paying for the noise margin can be transmitted, he noted, adding “There’s a communication problem. We have a sender as one of the chips, we have a receiver as the other chip. And there is noise in this channel.”
“Our job as communication experts is to design a set of permitted signals and a method to detect them in such a way that the system is resilient to noise,” he highlighted.
Shokrollahi further talked about detectors, primarily pointing to Differential Signaling and Differential Amplifier circuit.
Differential Amplifier circuit needs to stay close to current technologies, he maintained, stating that no exotic designs should be used, as they would most likely not work in practice.
He introduced a new modulation scheme for chip-to-chip communication called “chordal codes.”
These codes are somewhat reminiscent of spatial MIMO systems and provide the first step toward better utilization of the available communication bandwidth between chips.
“In Chord Signaling, we use generalized differential amplifier circuits. They are easy to implement. It is not exotic and it is robust in implementation,” he noted.
Following this, they tried to “abstract things into a mathematical framework, solve the problem in that framework, and go back to the task,” he said.
Shokrollahi delineated the mathematical abstraction in terms of values on the wires and detectors, for both of which there are set of vectors.
Pointing to “detector computes sign of inner products,” he said it can be done with standard electronic circuits.
“Now that we have a mathematical abstraction, the task is to design these two sets of vectors such that communication is resilient to noise, hence lower power,” he noted.
Explaining their theory and practice, he said they have drawn upon discrete mathematics, probability theory, statistics, information theory, embedded algorithms, design and implementation of analog and digital circuits, along with a lot of customer conversations.
Shokrollahi said that in 2010, they started to develop the theory and patents of Kandou company at EPFL. Then, they tried to “find a cool name, developed a prototype, and prepared to spinoff.”
He introduced the members of initial Kandou team to the audience and talked about their fun times as well as their hard times in Kandou.
Giving advice to those who aspire to have an academic startup, he said “first you need to have a technical idea based on multiple years of research.”
“Then you want to bring this idea to the market,” he said. Shokrollahi compared this stage to “slingshot around a black hole to gather momentum.”
“As we are academicians, the black hole is our non-knowledge about what the market is. Very often you just drop into the black hole. There are rare cases where you compute everything right and if you are lucky, you can slingshot around the black hole and start to have a company,” he continued.
“However, a nonacademic who actually comes from the market knows exactly how to slingshot around a black hole,” he maintained.
According to Shokrollahi, when they started the Kandou company, they had “great technology, vision, and enthusiasm, but had no focus on products, no market intelligence, no top down market approach, and no real business model.”
He further talked about their spinoff and growth, saying “we spun off the Kandou company out of EPFL, and we used the excellent ecosystem of Switzerland to fund the company for the first year. Then we went on raising fund for many years from reputable companies for the idea that we had.”
“I think we have raised $207m funding to date,” he stated.
The 2017 Mustafa
Prize laureate then recounted the story of fluctuations in their enthusiasm at Kandou company throughout the years.
The toughest days included mainly the time period when enthusiasm of the team was “dangerously low,” he noted, adding “This was exactly the time when we were trying hard to market the fundamental technology that we had, and went through lots of failures.”
“We passed the tough days and now we are bigger,” Shokrollahi said.
They have headquarters in Switzerland, England, Scotland, Germany, Denmark, Taiwan, and other countries.
They are active in several areas: chiplets, satellites, image sensors, USB 4, and PCIe.
“Now we are more than 200 people in Kandou. We have 8 offices worldwide and we have more than 500 patents,” he told the audience.
“If you ask me where we are now considering the black hole metaphor I mentioned earlier, the way I see it is that we are still spinning into the unknown, but maybe around the unknown at this point,” he concluded his speech.
The 2017 Mustafa
Prize in Information Theory was awarded to M. Amin Shokrollahi in recognition of his outstanding work on Communications: Raptor Codes. The world’s most advanced forward error correction (FEC) code for data networks, Raptor codes invented by Shokrollahi in 2001, provide protection against packet loss by sending additional repair data used to reconstruct “erased” or “lost” data.
