Introduction
He is the director of the Dedicated Processor Research Institute at Beijing Institute of Technology, a lifetime chair professor at Linköping University in Sweden, and a senior member of the IEEE; he previously served as the chief engineer in the integrated circuit R&D department at Ericsson and successfully founded two dedicated processor companies in Sweden. He is Liu Dake, an expert from the national ‘Thousand Talents Program’. Today, let Professor Liu take us on a journey through the technological world of high-end embedded systems.
With the strong wave of the mobile internet, embedded systems, which are increasingly popular alongside mobile smart terminals in the information technology field, are becoming more prominent. These systems are application-centric, based on computer technology, and can be tailored in software and hardware to meet the strict requirements of application systems in terms of functionality, reliability, cost, size, and power consumption.
Although embedded systems have only recently received special attention, they have actually been developing for 30 years. The microcontroller (a type of integrated circuit chip) from the 1970s was a simple embedded system. Today, with the popularization of various cutting-edge concepts such as the Internet of Vehicles and the integration of three networks, related technologies and products have been developed, and high-end embedded systems have entered our daily lives. At the same time, the ever-expanding computational demands have also posed higher requirements on embedded systems, presenting them with challenges.
Liu Dake, an expert from the national ‘Thousand Talents Program’ and director of the Dedicated Processor Research Institute at Beijing Institute of Technology, believes that to continuously meet future demands, embedded systems need to start by updating the core components of this value chain. Liu predicts that dedicated processors will replace the currently mainstream non-programmable dedicated integrated circuits as the core of high-end embedded systems.
To this end, we conducted an exclusive interview with Liu Dake to elaborate on various reasons.
Dedicated Processors ‘Debut’
Reporter: For most people, the concept of embedded systems may still be somewhat unfamiliar. Could you please explain it briefly?
Liu Dake: Simply put, an embedded system is any dedicated computing system other than general-purpose computer products. For example, a simple embedded system could be the control system of a refrigerator or microwave oven, while a more complex one could be the motor frequency control system in an air conditioner. The most typical high-end embedded system is the computing system in a mobile phone.
Now, with the development of related technologies, complex devices such as cars, medical devices, and robots will all be connected to the internet, and they may contain several embedded systems internally. Compared to simpler systems, these embedded systems will be more advanced and complex, with increasing demands for computational power and power consumption. For instance, a baseband chip for three-network integration or a baseband chip for the Internet of Vehicles requires computational capabilities of nearly 100 billion operations per second, with power consumption around one watt; the requirements for 4G mobile phone communication systems are even higher, with computational capabilities nearing 100 billion operations per second and power consumption needing to be in the hundreds of milliwatts, which is equivalent to dozens of times the computational power of a regular computer, while the power consumption is only a fraction of it.
Reporter: In your view, what challenges and difficulties do embedded systems face today? How can they be resolved?
Liu Dake: As mentioned earlier, future products will increasingly demand higher computational power and lower power consumption from systems, and the computational power and power consumption of a system often depend on the embedded processor within it, which is the core computing and control component of the embedded system. Since general-purpose processors must cater to all possible applications, they cannot simultaneously provide high performance and low power consumption. Therefore, most mainstream high-end embedded systems currently do not use processors but instead use non-programmable dedicated integrated circuits tailored for specific users to replace their functions. The basic architecture of dedicated integrated circuits uses microcontrollers with non-programmable accelerators, which lack flexibility, especially in meeting the requirements of various protocol standards.
Taking mobile communication as an example, future mobile phones will need to handle two 4G standards, several 3G and 2G standards, and wireless LAN standards. Clearly, non-programmable dedicated integrated circuits will struggle to meet these requirements. Additionally, the design cycle for non-programmable dedicated integrated circuits is long, and their product lifespan is short; sometimes, just as experimental chips are completed, they become outdated before mass production, necessitating improvements, with each improvement requiring a basic investment of hundreds of millions of RMB. Looking at it this way, the costs of this solution are extremely high, and most companies are already finding it difficult to bear. Therefore, I believe that dedicated integrated circuit solutions for high-end embedded systems are likely to ‘expire’ before 2020.
Thus, traditional processor architectures are facing significant issues, and embedded systems, especially high-end embedded systems, must seek new paths, with the best path being the use of dedicated processors, as their design intent and methods are suitable for high-end embedded systems.
Reporter: So, what is a dedicated processor?
Liu Dake: The English abbreviation for dedicated processors is ASIP, which stands for Application Specific Instruction-set Processor. Academically, it is a specialized instruction set architecture designed for a specific application domain, possessing programmable flexibility within a specific domain, and it can improve its design through software updates.
The concept of dedicated processors is not unfamiliar; however, the terminology varies in academia. For instance, Germany had the MIMOLA dedicated processor automatic synthesis system in the 1970s. Research on this topic has been ongoing in foreign academia, with statistics showing that thousands of related academic papers are published each year. As the drawbacks of non-programmable dedicated integrated circuits become increasingly apparent, research on dedicated processors has gained more attention, gradually forming a school of thought, although China is somewhat lagging in this area academically.
In 2011, the then director of the Electronics Department of the Ministry of Industry and Information Technology, Xiao Hua, emphasized at the National Semiconductor Annual Conference the need to vigorously develop application-oriented personalized processors, which is essentially what we now refer to as dedicated processors.
Five Advantages of Dedicated Processors
Reporter: Why do you think they can replace dedicated integrated circuits and become the computing platform for future embedded systems?
Liu Dake: Dedicated processors possess the high performance of non-programmable integrated circuits ‘oriented towards applications’ while also having the flexibility of general-purpose processors ‘programmable’. This is because dedicated processors have several characteristics that allow them to not only compensate for the shortcomings of both general-purpose processors and dedicated integrated circuits but also save costs and improve efficiency.
First, dedicated processors can provide extremely strong computational capabilities, better fulfilling core functions. Because they are designed for specific fields, they have low redundancy and high energy efficiency, with computational capabilities generally dozens of times that of general-purpose processors. For example, a versatile athlete may find it difficult to break a high jump record; that is the job of a specialized high jumper.
Second, dedicated processors can achieve ultra-low power consumption. In embedded systems, power consumption mainly comes from data movement, and the data movement of dedicated processors is designed for specific applications, making it more targeted, thus naturally resulting in lower power consumption.
The third point is that dedicated processors have strong flexibility. As mentioned earlier, dedicated processors can be flexibly programmed within specific application ranges, allowing them to achieve compatibility with multiple standards. In contrast, general-purpose processors must meet all possible user requirements under any conditions, which makes their flexibility much weaker.
The fourth advantage of dedicated processors is that they can reduce silicon area costs through hardware reuse. High-end multi-standard non-programmable integrated circuits are typically assembled from multiple modules without reuse, while dedicated processors can run different software. In 2006, our team provided a multi-standard UMA baseband dedicated processor chip to the United States, with an area of 5 square millimeters, while a competitor using a non-programmable solution without hardware reuse had a chip area of 13 square millimeters.
The fifth point is that dedicated processors have shorter design cycles. Generally, a common method to shorten design cycles is to separate hardware and software and then develop them in parallel. The design principle of dedicated processors follows this approach, so their design cycles are naturally shorter.
Reporter: You mentioned that research on dedicated processors has been ongoing for many years in academia. What are the main research topics in this area now?
Liu Dake: Research on dedicated processors adds three layers on top of integrated circuit research.
First, there is an analysis of computational characteristics and acceleration needs, including analysis tools and application cases, aimed at obtaining the instruction set architecture for dedicated processors. Second, there is microarchitecture design, which includes designing processor hardware code, parallel programming, and assembly programming tools. Finally, there is the evaluation of dedicated processors, which involves testing capabilities, estimating power consumption, and estimating costs.
Dedicated Processors in China
Reporter: What areas does you and your team mainly focus on?
Liu Dake: Our research mainly focuses on microarchitecture design and acceleration analysis, specifically on the automation and optimization of dedicated processor design. Automation aims to establish a set of design processes and tools for dedicated processors to minimize design time while stimulating engineers’ creativity in design. Optimization of dedicated processors involves researching their structure and performance to minimize redundancy and maximize performance.
In this area, we have already achieved some industrial results. In April of this year, we sold a baseband processor aimed at wireless broadband terminals and wired broadband terminals for 230 million RMB plus 30 million RMB in options, which is known as a software-defined radio and a supercomputing platform for three-network integration. Additionally, we have two projects nearing completion: one is a handheld embedded supercomputing platform for communication, multimedia, and 3D video game consoles, and the other is a processor automatic synthesis system.
Reporter: However, you previously mentioned that overall, China is still somewhat lagging in this area. What does this mainly refer to?
Liu Dake: China’s research in this area started relatively late. Dedicated processors are core components for equipment manufacturers, who use them to achieve high-performance computing while also protecting them as core secrets. Before returning to China, I worked for many years in European companies and am well aware that foreign companies tend to keep their research progress in this area under wraps. For example, Ericsson’s base station dedicated processor has a history of twenty years, but very few people in China know about it. In recent years, with increased exchanges with foreign companies, we have learned that top international equipment manufacturers possess many core chips, but it is difficult to discern which are non-programmable dedicated integrated circuits and which are dedicated processors.
Moreover, large foreign equipment manufacturers have first-class research institutes or core design departments, but few domestic companies have such strategic technology research institutes. Core technologies are often held by research institutes under various ministries, but companies cannot hand over core secrets such as algorithms in the system to research institutes for the development of dedicated processors. This adds extra difficulty to research in this area.
However, in recent years, I have also seen some scattered research activities in China, but these studies have not yet been integrated into a dedicated processor research system. This is also the biggest gap between us and foreign countries.
Reporter: In your view, what do you think is the biggest obstacle to catching up or even surpassing in this area?
Liu Dake: Research and industrialization of dedicated processors are actually based on research into general-purpose processors and traditional integrated circuits. So, strictly speaking, our foundation is not poor, and our capabilities are not an issue; we could even be on the same starting line as developed countries.
However, compared to developed countries like the US, Japan, and Western Europe, we have always had a weak ability to transform quickly and a lack of keen insight and rapid follow-up capabilities for new technologies. I am currently very concerned that the government’s deployment and follow-up in this area are too slow, which may cause us to fall further behind other countries.
Government Should Act Promptly
Reporter: What do you think relevant government departments should do specifically?
Liu Dake: As mentioned earlier, the market for dedicated integrated circuits will continue to shrink in high-end embedded systems, and general-purpose processors cannot achieve high performance and low power consumption simultaneously, making dedicated processors the best choice. It should be said that the national level has already recognized this point, and relevant leaders from the Ministry of Industry and Information Technology have clearly stated in public that they need to vigorously develop application-oriented personalized processors. However, the key now is to deploy promptly.
Specifically, the government should establish corresponding research projects to cultivate a large number of related talents in the long term, especially high-end talents such as PhDs and Masters in dedicated processors, while also accumulating patents in this area. This way, we can ensure that we do not fall behind in intellectual property in international competition and can even use it as an advantage. Additionally, the government can encourage large enterprises, such as China Electronics Corporation, CETC, and Datang, to establish dedicated processor R&D teams or even set up corresponding subsidiaries, which will give us more confidence in the industry.
Reporter: Do you have any other suggestions?
Liu Dake: Dedicated processor companies require significant initial investment, while China’s venture capital culture has not yet kept pace. Due to low investment returns and long cycles, dedicated processor investments do not appeal to venture capital firms. They are generally reluctant to invest in such projects, which is difficult to change in the short term. Therefore, investment in dedicated processors needs to rely on the government. The state should increase initial investments in dedicated processors; only with government leadership can we make things bigger and better.
In fact, as mentioned earlier, due to the rapid development of the computer industry, dedicated integrated circuits often need design updates before they are even produced, which incurs particularly high costs. In contrast, dedicated processors can avoid modifications to chip designs by changing software, significantly reducing investment. Therefore, compared to the investment in dedicated integrated circuit companies, the investment in dedicated processor companies is much lower. In this regard, the government may achieve higher returns with relatively lower costs in dedicated processor investments.