A paper on the dispute over the architecture of the IoT master chip

The ZTE incident caused a global sensation, and everyone's eyes gathered on the lack of server, computer, and storage underlying chip technology. Shares of domestic chip suppliers such as Ziguang rose.

The incident reflected that China is weak in the chip and its industrial chain; after all, in the PC era, we started too late. However, in the era of the Internet of Things (chip), we still have the hope of overtaking the corner. As one of the important parts of the Internet of Everything, the Internet of Things chip includes chip industries such as security chips, mobile payment chips, communication RF chips and identification chips. It is estimated that the scale of China's Internet of Things will reach 1.5 trillion in 2020. Next, this article mainly introduces several architectures of the IoT master chip.

Domestic and foreign giants have laid out IoT chips

The bright market prospects of the Internet of Things and the unfinished IoT master chip architecture market have attracted domestic and foreign giants to exert their strengths and seize the commanding heights.

In foreign countries, Intel released the x86-based Edison chip in 2014, followed by the introduction of the x86-based Curie chip in 2015. Qualcomm naturally refuses to stagnate in the mobile field and is based in 2016. The Krait300 architecture is based on the Snapdragon 600E and 410E IoT chips. The Krait300 architecture is based on the ARM V7 instruction set. The performance is between the ARM-designed A9 and A15 architectures. Samsung also released Artik1, 5, and 10 in 2015. Networking chips are based on the ARM architecture. In addition, giants such as Google, AMD, and Nvidia have also developed IoT chips.

In the domestic market, MediaTek launched the MT2503 based on ARM v7 architecture in 2015. It has been widely used in the field of shared bicycles. It has reached an agreement with Microsoft this year to jointly launch the first AzureSphere chip MT3620; Huawei Hisilicon in September 2016 Launched the first official commercial IoT chip, and its Boudica 120 and 150 chips were also shipped in large quantities in the second half of 2017, all based on ARM architecture. In addition, SMIC, Hua Hong Hongli, TSMC, Spreadtrum, China Resources Micro, and United Manufacturers such as Core Technology have also laid out the IoT chip market.

IoT chip architecture

The premise of the Internet of Everything is the connection between the intelligent terminal device and the sensor. Its application scenarios and characteristics make the IoT chip biased towards low power consumption and high integration. Low power consumption enables developers to add functions to power-constrained devices while maintaining the chip. Size, expand the application possibilities. Adding highly integrated components enables plug-and-play of the chip, simplifies application development, facilitates device replacement, and facilitates rapid product introduction.

Due to the characteristics of IoT applications and the needs of the scene, efficient, streamlined instruction sets and low-power chips are better choices. So what architecture will be suitable for IoT-specific chips?

ARM architecture

With the development of intelligent machines, the ARM architecture has become a great achievement in the past few years. ARM is one of the representative works of RISC microprocessors. The biggest feature is energy saving, widely used in embedded system design, and even many people’s minds. The default IoT chip is the ARM architecture.

At present, the chip architecture on the market is mainly based on x86 and ARM. Compared with the x86 architecture based on complex instruction set, the ARM architecture uses a reduced instruction set, which is more compact and consumes less power. The characteristics and application scenarios of the Internet of Things require that the chips used must consider power consumption and integration. This makes the IoT chip based on the ARM architecture occupy an innate advantage in the era of the Internet of Everything.

This is also the case. The Qualcomm 600E and 410E IoT chips mentioned above, the Huawei Boudica 120 and 150 IoT chips, and the Samsung Artik1, 5, and 10 IoT chips are all based on the ARM architecture. MediaTek uses the ARM v7 architecture. The MT2503 IoT chip has been widely used in the shared bicycle field.

X86

The IoT chip based on the x86 architecture in the market is probably only Intel. Its Edison and Curie launched in 2014 and 2015 are based on the x86 architecture. Perhaps it is the helplessness of the x86 architecture in the Internet of Things era. Intel's subsequent chips for the Internet of Things are still based on the x86 architecture, but internally streamlining system instructions and controlling energy consumption issues, this approach also disguised the x86 architecture. In the era of the Internet of Things, it will not work.

So Intel announced a $16.7 billion acquisition of Altera (Altera), the world's second-largest FPGA vendor, for about $54 a share, the largest acquisition in the history of Intel's 47-year history. The acquisition of this Intel corresponds to a valuation of 35 times, which is very rare in the semiconductor field.

Intel's acquisition of Altera is based on three considerations:

First, IBM is working with Xilinx, the world's largest FPGA vendor, to focus on big data and cloud computing, which has caused Intel's huge concerns. Intel has fallen behind in mobile processors, and big data and cloud computing can no longer lag behind. Intel will cooperate with Altera to open the internal interface of the Intel processor to form a combined mode of CPU+FPGA. Among them, FPGA is used for shaping calculation, cpu performs floating point calculation and scheduling, and this combination has higher unit power performance and lower latency.

Second, IC design and filming costs. As the semiconductor process index grows, FPGAs will replace ASICs in high-value, relatively small-volume (less than 50,000), multi-channel computing in the IoT space. At the same time, the FPGA development cycle is 50% shorter than the ASIC and can be used to quickly seize the market.

Third, Intel's acquisition of Altera is actually a good opportunity in the field of Internet of Things. One of the main functions of FPGAs before was to use for prototyping. First, use FPGA to perform functional verification and then use ASIC to stream, in order to better design ASIC in the case of cost saving. However, with the performance and capabilities of the FPGA itself and the complexity of the achievable logic, FPGAs can now directly replace some of the ASICs and DSPs in high-performance, multi-channel computing. The main reason is three: parallel computing The hardware structure is variable and can be changed during operation.

FPGA

One of the characteristics of the Internet of Things is the diversity of scene applications. This requires that the functions of the IoT chip do not have to be complex, but need to meet the development needs of different scenarios. The emergence of FPGAs provides new possibilities for IoT chips. Its programmable design advantages can help developers shorten development time and adapt to the requirements of algorithms from hardware structure changes to meet the customization needs of IoT diversity.

FPGAs are mainly used in video, industrial, communications, automotive and other fields, because these areas are generally characterized by low chip usage, severe application differentiation, incomplete functions, low power requirements, and stable and reliable performance requirements. But for consumer applications, the primary consideration is price and functionality.

Cheap FPGAs are several times higher than other chips. For example, xilinx FPGAs cost at least $10. In addition to price, FPGA power consumption is at least an order of magnitude higher than that of the main control chip. It is almost impossible to apply in the field of wearable devices, and is used in some high-end consumer market applications, such as high-end TV sets. Central air conditioning, etc.

The attitude of international manufacturers is basically not optimistic about the application of FPGA in consumer intelligent hardware. Price and power consumption are the biggest challenges.

Although power consumption and price are not dominant, the advantages of FPGA-programmable and port-flexible are especially important in the case of increased customization requirements.

RISC-V

In addition to ARM, another architecture based on a reduced instruction set is RISC-V.

RISC-V is an open source instruction set architecture (ISA) based on the reduced instruction set by Yuan Ze. Since the beginning of the RISC-V architecture in 2010, it is growing rapidly with the advantage of open source (free). RISC-V-based chip companies continue to emerge, opening up to the Internet of Things, artificial intelligence, embedded markets, and the closed commercial ISA has shown great vitality.

The RISC-V instruction set is designed with a small, fast, low-power, real-world implementation in mind, but without the over-architecture of a particular microarchitectural style. In addition, another major advantage of RISC-V is open source, which means that manufacturers do not need to spend money to buy ip, modularity allows users to freely customize according to their needs, configure different subsets of instructions.

It is understood that Tesla has joined the RISC-V Foundation and is considering using the RISC-V design in the new chip. Western Digital and NVIDIA have announced plans to introduce this new chip design into some products.

Although it is still difficult to challenge ARM with the RISC-V architecture, which is only 8 years old, it provides a new way of thinking and direction. Faced with the rapid growth of the Internet of Things platform and the innumerable growth of the Internet of Things application, the open source RISC-V architecture will bring lower cost chips, which will certainly attract some cost-sensitive IOT companies with better flexibility. It also brings the possibility of customizing the chip.

Of course, RISC-V is still only a microprocessor, and it takes longer for Unix-level application processors to implement applications. In terms of the current market, Intel dominates the data center field, and ARM dominates the mobile space. For the upcoming Internet of Things era, the ecological chain is still in an emerging state, and lower cost and more free design applications may face a big outbreak for RISC-V.

ASIC

The advantage of ASIC chip is that it has strong computing power and low mass production cost. However, the development cycle is long and the cost of single-stream is high. It is mainly applicable to fields with large quantity, high computing power requirements and long development cycle, such as most Consumer electronics chips and experiments.

Compared with ASIC, the power consumption of FPGA is relatively large, and the cost advantage is not enough to support the cost-effective IoT. Therefore, the current application of zui for edge cloud computing is more selective.

ASICs can be customized into different integrated circuits according to different product requirements. Although ASICs have higher performance and lower power consumption than FPGAs, their flexibility is not as good as FPGA. For example, at the end of July this year, Google released the ASIC chip Edge TPU, which aims to complement the cloud cloud TPU chip and further improve the Internet of Things market.

to sum up

The Internet of Things chip has become the scent of the giants at home and abroad. The Intel x86 architecture dominates the PC chip for 20 years. The ARM architecture dominates the mobile communication chip. Has the ARM or SD architecture continued to flourish in the Internet of Things era?

ZTE's lifting of the ban shows that the global electronics industry is an ecology. It is unrealistic to attempt to control the entire ecology by controlling only one link, such as chips. And because it is an ecosystem, it is also controlled and restricted by each other. In addition, the demand for chips varies according to the application field, and no one can beat the world.

At present, it is necessary to analyze and layout the Internet of Things technology from the perspective of overall ecology, and it is necessary to strengthen its own strength and create barriers to innovation in key technologies.