Zhou Hong, President of Huawei Strategic Research Institute: Future-oriented scientific assumptions and business visions

At the 19th Huawei Global Analyst Conference held today, Zhou Hong, President of Huawei Strategic Research Institute, delivered a speech on "Scientific Assumptions and Business Visions for the Future". He said that Huawei is creating with partners around the world with an open mind. On April 30, Huawei plans to launch the online Huang Danian Tea House, hoping to build a channel for scientific and technological exchanges and open to the whole society. At the Huang Danian Tea House, Huawei will summarize and refine the challenges and invite outstanding talents from around the world to explore innovation together.

The following is the full text of the speech:

Hello ladies and gentlemen! Welcome to the 19th Analyst Conference.

We are pleased to have the opportunity to discuss future-proof scientific hypotheses and business visions.

We know that the 18th century was the era of mechanization, the 19th century was the era of electrification, and the 20th century was the era of information technology, so what era will the 21st century be?

I believe that the 21st century will be an era in which human society is fully intelligent, and the core of intelligence is perception, connection and computing, and the resulting higher cognition and control of matter and phenomena, life and energy.

On the way to the intelligent world, we are facing great challenges, on the one hand, happy life, efficient work, green environment also need to perceive, connect and calculate to improve hundreds of times the ability,

On the other hand, in related science and technology, there has been no major breakthrough in the past few decades, and it has even approached the bottleneck.

How can we create a viable development route?

I believe that facing the future, only by boldly proposing hypotheses, boldly proposing visions, and daring to break the rules and regulations of existing theoretical and technical bottlenecks can we make great strides forward.

Digital technology has greatly enriched the work and life of human beings

In the past 10 years, with the rapid development of broadband communications, smart devices, AI and cloud computing, digital technology has greatly enriched people's lives.

From phone calls, browsing information online, instant messaging, to map navigation, e-banking, online shopping, etc., ICT technology has become an indispensable and important part of more and more people's lives.

In addition to the ICT technologies needed in life, we have also conducted joint explorations with many industries in the past 10 years to see if ICT technologies can help or how to help the industry develop.

For example, we work with car companies, telecom operators, to conduct experiments on highways,

As we all know, the response speed of manual driving is measured in seconds, and we help to increase the detection and response capabilities of emergencies by hundreds of times by establishing a high-performance connection of 10 milliseconds between cars and between cars and networks.

At a speed of 100 kilometers per hour, the safety distance of tens of meters and hundreds of meters required for manual driving can be reduced to 0.8 meters, thereby greatly improving the traffic flow and safety of the highway.

At the same time, it can support team driving to reduce wind resistance and save about 20% of fuel consumption.

The Internet of Vehicles can also support remote driving, creating new operating modes and service models.

We are also experimenting in urban environments, and through the Internet of Vehicles and vehicle-road coordination that exceed the visual distance, it is possible to improve traffic efficiency by 30% and reduce traffic accidents by 90%.

Today, in factories, hospitals, ports, and coal mines, ICT technology is penetrating into thousands of industries, enabling the digital and intelligent transformation of the industry.

Humanity's quest for the future is never-ending

Looking to the future, we see that there are still many places where ICT technology has the potential to make a greater contribution.

For example, human health and well-being can better support the management of sports health and chronic diseases through wearable sensors, wireless communications and cloud computing, and AI computing can also help with the rapid design and efficient screening of drugs and vaccines.

ICT technology can support the ubiquity of autonomous and intelligent machines, thereby improving people's quality of life and improving the efficiency of operations in all walks of life.

ICT technology can help build a green and sustainable environment, such as efficient energy conversion and dispatch, design of low-cost, high-efficiency energy conversion catalysts, energy storage materials.

In the digital world where virtual and real are integrated, ICT technology can also help establish an experience that is "far away, close at hand, and immersive", enrich people's lives, help people learn and grow, and help all walks of life to rapidly iterate and improve in the digital world.

Global digitalization surpasses the development of "ten hundred times"

Driven by so much demand, global digitalization is growing at an exponential rate.

For example, mobile broadband data traffic around the world has grown from 0.24 exabytes (exabytes) per month in 2010 to 60 exabytes per month in 2020, an increase of more than 250 times in 10 years.

China's mobile broadband data traffic has grown from 0.033 exabytes per month in 2010 to 13 exabytes per month in 2020, an increase of more than 400 times.

Looking to the future, we believe that digital technologies will grow at a rate of more than a hundred times in more than a decade, and digitalization will promote the accelerated development of people and society.

On the other hand, we also see that many existing theories and technologies were proposed decades or even more than a hundred years ago, and the application of these theories and technologies has begun to encounter bottlenecks, such as Nyquist sampling theorem and Shannon's law in the field of communication, the theory of computability in the field of computing and the von Neumann architecture, Moore's law in the field of semiconductors, etc., hoping to have new assumptions and visions to drive breakthroughs.

To this end, we propose four scientific hypotheses and business visions for the future, and hope to explore together with academia and industry to carry out future-oriented research.

One: Expand the boundaries of cognition, matter and energy, phenomena and laws

The first is to explore basic science and cutting-edge technologies to expand the boundaries of our cognition. In particular, breakthroughs in physics, chemistry, biology and other fields will enable us to better invent new molecules, catalysts, proteins and other materials and devices, as well as new equipment and new processes.

Once, I was discussing with a quantum scientist, how to save photons and quanta? When he proposed the concept of quantum storage in 1993, no one believed it, and everyone might think, can you use a bottle to store light? Does the operation of storing a quantum affect its state? Until 1998, Harvard Hau et al. used electromagnetic induction transparency to reduce the photon speed to 17 m/s, and in 2000, they successfully "frozen" the photon for a minute. In 2006, Imperial College's Pendry et al. proposed that light could be bound by a "photon black hole" approach that could be used to keep it from leaving. There are already many ways to achieve quantum storage to better support quantum communication and quantum computing.

In order to reduce the power consumption and improve the reliability of semiconductor devices, we work with scientists to analyze the thermal mechanism in semiconductor devices to see if favorable conditions can be constructed to accelerate the transformation of "photoacons" into "sonons", thereby reducing the formation of hot spots between the gate and the drain.

Now that many superconducting quantum computers use millikelvin temperatures, some scientists are exploring further, using lasers to cool atoms, lowering the temperature from Haucelvin by a million times to Nagarvine, approaching the temperature limit of absolute zero, and see if more complex quantum phenomena can be found.

In the future, can the characteristics of matter be predicted by calculation without relying on lengthy experiments? The answer is possible. For example, using the USPEX calculation method, the main properties of molecules composed of less than 200 atoms can be calculated with a computing power of 1 million nuclei. In 2017, scientists discovered the structure of superhard tungsten pentaboride through calculations, solving the problem that has plagued the scientific community for nearly 60 years; in 2019, scientists found that thorium ten hydride has amazing high temperature superconductivity at a pressure of 850,000 atmospheres, and the critical temperature reaches -112 degrees Celsius.

With better computational chemistry, we can expect to discover or invent better catalysts, chemicals, biologicals, and vaccines.

Two: Expand the limits of perception and better understand the world and humanity itself

The second is that in the future we will continue to expand our ability to perceive the world and perceive ourselves, from approaching human perception to transcending human perception, from alternative perception to extending and creating perception, from human perception to machine perception.

In this regard, we need to learn from the biological community, and nature has evolved over millions or even hundreds of millions of years to form a perception that far exceeds that of existing machines and humans.

For example, visually, some spider eyes far exceed the human eye in object contour and motion calculation, which is conducive to quick and accurate capture of prey, I wonder if self-driving cars need this kind of eye?

The same is the frog eye, which is a highly sensitive single-photon receiver that can see more clearly in dark environments.

In terms of smell, the dog's nose is more capable of distinguishing odors than humans by 1,000 times.

In addition to expanding our perception of the external world, we will be able to better perceive and control the human body itself in the future. Technologies such as ECG, EEG, PPG, etc. have not yet been systematically, conveniently and at low cost, and we still have a lot of work to do for the real-time measurement perception of the eight subsystems of the human body. Through the development of new sensors, we may in the future measure blood pressure, blood sugar, ECG and other important health parameters in real time and without perception; we can develop new nervous system brain-computer interfaces, muscle-computer interfaces, better coordinate with machines, and in the future it is possible to use thinking to communicate and work, and use thinking to drive and play.

We can also develop new experiences that blend virtual and real into the digital world, such as 3D displays and virtual haptics, to help "see and feel" in the digital world.

Three: Explore new computing models and implementation methods, understand the world, and solve problems

The third is to explore computing models and efficient implementation methods that adapt to the goals and environment, so as to better understand the world, solve problems, and create value.

After years of accumulation in the information field, more than a dozen widely used computing modes have been developed, such as the butterfly computing mode based on fast Fourier transform in wireless and optical communications, the finite state machine computing mode based on logical state transfer in routers, and the statistics-based and related computing mode in AI. Mathematicians and engineers have struggled for so many years, have we come to an end in computational models? I think there's still a lot of room, for example:

In communication: as the future communication system continues to move towards high frequency and high speed, we will face more and more problems caused by nonlinear channels and nonlinear devices, can we expand from the traditional linear Fourier transform to the nonlinear inverse scattering transformation to better match future applications?

On AI: As applications continue to expand, we face the problem that statistically related AI computing patterns are unexplainable and unprogrammable, and there are also great energy efficiency challenges. Can we learn from the biological world, such as ants, small ant brains generally only consume 0.2 milliwatts of energy, it does not need to use deep learning, nor does it need to follow the theory of computability and the von Neumann architecture, but it can run around and do a lot of complex things, such as nesting, finding food, raising aphids and so on. Current self-driving cars also need tens of watts or even hundreds of watts to calculate, and there is still a big gap in energy efficiency compared to ants. Therefore, in the field of AI, in addition to statistics and related calculation modes, can we further develop mathematical logic calculation mode, geometric manifold calculation mode, game calculation mode, etc.?

In scientific computing: we use a lot of matrices, for the multiplication of two n-row n-column matrices, if according to the original simple algorithm, the complexity is n to the 3rd power, the Strassen algorithm created by German mathematicians in 1969 reduces the complexity to the 2.807 power of n, and at the end of 2020, Williams of MIT and Alman of Harvard gave an algorithm with a complexity of 2.3728596 to the power of n.

In matrix computations, we are more concerned with solving sparse linear equations, because in the social sciences, there are billions of people on Earth, and the average person maintains no more than 200 valid relationships; in chip design, the constraints of most components are local. In this field, Peng Yang and others at Georgia Tech invented an advanced algorithm to the power of n to the power of n, which won the 2021 Best Paper Award from SODA, the top meeting on computational theory. A few months ago our mathematicians invented a newer algorithm that reduces the complexity to the 2.28 power of n, which is 0.0516 to the power of Peng Yang et al., what does this progress mean? For n= 1 million, the computational complexity will be further reduced by about 45%.

In terms of specific implementation, supercomputers often use huge energy consumption to achieve large computing power, such as 30 million watts to achieve nearly 500PFLOPS computing power, while the human brain can achieve nearly 30PFLOPS with about 20W, and the efficiency is about 80,000 times higher.

From this perspective, should we develop adaptive and efficient computing models and create new architectures and components without being constrained by traditional computability theories and von Neumann architectures?

Four: Break through the assumptions of Shannon's law and develop information communication in a larger space-time

The fourth is to explore information communication in a hypothesis that differs from Shannon's law and in a larger space-time space, thereby overcoming spatial barriers and building global direct access, connecting the virtual and real worlds, and ubiquitous machines.

In the future, if the real-life holographic communication does not compress the data, it needs a bandwidth close to 2Tbps, and a delay of 1-5ms;

Autonomous driving, with 12 cameras, could generate up to 4 Tbytes of data per day, and current 5G networks are far from reaching this capacity.

Do we have enough theory and technology to achieve these challenges? I think it's possible.

Theoretically, for example, if we assume that the world has a priori knowledge and memory, we might jump out of the limits of Shannon's 1/2/3 law. In engineering, a quantum cascaded laser can generate hundreds of wavelengths at the same time, achieving hundreds of T of traffic; in the future, if we can make a high-heavy frequency attosecond laser, it may even generate millions of T of traffic. If these technologies can be grafted into the wireless and optical fields, can they improve communication performance thousands of times?

Open up scientific hypotheses and business visions to create knowledge and value

In order to open up scientific hypotheses and business visions, we divide innovation into five related links: from assumptions and visions to theoretical, technical and business innovation.

The closer you are to the back-end business, customers, and users of innovation, the more noticeable the effect, and the closer you get to the front-end assumptions, vision, and basic science, the more patience is required.

Facing the future, we must dare to seek answers from front-end basic research.

In basic scientific research, in addition to supporting the "Bohr Quadrant" innovation driven by the interest of scientists, we hope to explore the "Pasteur Quadrant" innovation with our partners, which can both expand scientific understanding and create application value.

10 questions and challenges for the future

Focusing on the first four assumptions and visions, focusing on the Pasteur Quadrant, we have distilled two basic scientific issues that can be focused on in the future, as well as eight cutting-edge technology challenges.

The first scientific question is how do machines perceive the world, and can they build models that are appropriate for machines to understand the world?

The second scientific question is how to understand the physiological model of man, especially the operation mechanism of the eight subsystems of the human body, as well as human intention and intelligence?

Cutting-edge technology challenges include:

How to develop new perception and control capabilities on the human-machine interface, such as brain-computer and muscle-computer interface, 3D display, virtual touch, smell, taste and so on

How to continuously and perceptually measure a person's blood pressure, blood sugar, and ECG in health? Can AI and artificial intelligence help invent new chemicals, biologics and vaccines?

How to develop application-centric, value-oriented and experience-oriented, efficient automation and intelligent software in software?

How can you approach and extend shannon's limits in communications to achieve efficient, high-performance connectivity at the regional and global levels?

How can we develop adaptive and efficient computing models in computing, develop non-von Neumann computing architectures and non-traditional components, and develop interpretable and debuggable AI?

In terms of materials, how can AI help invent new molecules, catalysts and devices?

How to develop a technology that goes beyond traditional CMOS manufacturing in manufacturing to achieve lower cost and higher efficiency?

Can we develop safe and efficient energy conversion and storage in terms of energy and provide on-demand services?

The greatest force is synergy, and the strongest wisdom is the wisdom of the crowd

Huawei is creating with partners around the world with an open mind.

On April 30th, we plan to launch the online Wong Tai Nian Tea House, hoping to build a channel for scientific and technological exchanges and open to the whole society.

At the Huang Danian Tea House, we will summarize and refine the challenging topics and invite the world's outstanding talents to explore innovation together.

Today, all our imaginations about digital technology are conservative

Thirty years ago, when I was in college, I still needed to wait in long lines to make long-distance calls, and I couldn't imagine that one day I could hold a small box, and I couldn't communicate with my distant family anytime and anywhere without any connection, and through this little box I could connect the world and do a lot of things, which was too sci-fi at the time.

All our current imaginations of the future may be conservative, so we need to be braver,

I hope to work with academia and industry to reconstruct basic theories, reconstruct architecture, and reconstruct software, and jointly explore and create the future!

Thank you!