During the Chinese New Year, a piece of news did not attract the attention of most people. MPS announced the acquisition of Axign B., a start-up based in the Netherlandsv.The company specializes in programmable multi-core DSP (Digital Signal Processor) technology that delivers near-zero distortion signals for automotive and consumer audio systems while significantly reducing power consumption.
A while ago, Torch Core entered the field of automotive audio DSP chips, and laid out a wholly-owned subsidiary to inject capital into Shanghai Judi, Zhuhai Yixin plans to jointly invest in Shanghai Judi with Di Dingrui, Di Wei Sheng, and Rui Shengtai to help the research and development of DSP chips and technologies of Shanghai Judi Company, which also marks the official entry of Torch Core Technology into the field of automotive audio DSP chips.
A month earlier, AKM Asahi Kasei Microelectronics also launched a new automotive audio DSP.
With the increase in the demand for intelligent automobiles, audio DSP (ADSP) (ADSP) as a veteran is once again silently watched, and a dark war is coming.
Audio DSP, can you still eat?
I have to say that DSP is an "old thing", and now no matter what the field, this chip is rarely mentioned. But being old doesn't mean it's not important, and it doesn't mean it's been eliminated.
Unlike most chips on the market, DSP has its own "mission".
The world of watchmaking electronic engineering.
DSP is actually a "partial student", unlike the CPU, which can do any task, DSP is a microprocessor born in response to the needs of the times. Although "the sparrow is small and has all the organs", because the skill points are all focused on the architecture and instruction set design, DSP is good at various operations related to digital signals.
For example, the CPU is like a set of knives for a chef that can cut anything, and the DSP is like a paring knife that can only peel the skin, but the peeling efficiency is the highest.
There are three reasons for this divergence:
First, the DSP adopts the Harvard structure inside, in this structure, the program memory and the data memory use different buses, so it has a larger memory bandwidth, and the movement and exchange of data is more convenient, which is very suitable for processing digital signal tasks;
Second, DSP can complete tasks with higher performance and lower power consumption in a shorter time at a very low core frequency, which is very power-saving;
Third, DSP has a special hardware multiplier, such as single-cycle multiplication and addition instructions, reverse sequence addition and subtraction instructions, block repeat instructions, etc., and even a lot of commonly used multi-operation sequences are specially designed to maximize the work that can be completed by each clock cycle, which greatly improves the speed of digital signal processing.
So, to use the above analogy, if we have a better knife, should we give up the paring knife? The answer is clearly not.
This can be seen from the market data - the global audio DSP market size in 2019 is 110$600 million, which is expected to reach $234 by 2027$300 million, with a CAGR of 93%。From the perspective of terminal applications, mobile phones, smart homes, home entertainment systems, Internet of Things, and in-vehicle systems will be the main driving forces for the growth of DSP.
MCUs and SoCs encircle and suppress audio DSPs
With the premise that "audio DSP is definitely a very useful chip", and then look at the history of audio DSP, its living space has actually been squeezed.
In the early years, due to the limitations of floating-point performance, computing frequency, instruction set, etc., there was basically no better audio algorithm solution on the market except for DSP solutions. For a long time, DSPs, XPUs, and FPGAs have their own roles.
In 2010 and 2014, ARM launched Cortex-M4 and M7 one after another, and since then, some lightweight audio algorithms have begun to be directly put into the MCU. At that time, there were rumors among some developers in the industry that DSPs were about to be eliminated by ARM.
Time continues to advance, separate audio DSP chips are becoming more and more rare, manufacturers put in mobile phones and various devices SoC chips are very complex, including CPU, GPU, NPU, ISP and a series of cores, external DSP chips are also placed in this SoC. The goal pursued by manufacturers is to realize the functions of active noise cancellation, call noise reduction, transparency mode and tuning in a single device.
Whether it's an MCU or an SOC, squeezing out separate external audio DSPs, the market is less talked about about them.
After several generations, DSP currently exists in two main product forms: one is a single chip, that is, an external independent DSP chip, and the other is integrated in the SOC in the form of IP or processing unit.
Although DSP spends most of its time in obscurity computing and processing work, it still has excellent on-chip performance, an efficiently designed instruction set, rich audio interface resources, a mature and stable tool chain, and a deep upstream and downstream ecosystem for many years.
As Fang Jin, the former chief scientist of TI and the founder of DSP products, once said in 2012: "Some people may think that DSP as a product has gone from being worthless to creating billions of dollars a year and then disappearing." But this is really the beginning of a good news. It's not gone, it's just integrated into every digital processing system. The reason why we say this is because our efforts in IC technology have allowed the embedding of DSPs in chips. DSPs used to be very large, but now they are so small that they are almost invisible. ”
The story of audio DSP is never over, especially in the past two years, in order to enhance sound quality, the use of audio DSP has increased day by day, along with many more changes in line with the times:
Keep up with the trend of the AI era, combine with AI large models, and cover more audio scenarios, such as during CES 2024, GreenW**es Technologies launched the latest AI+DSP audio innovation technology, providing ultra-low latency adaptive active noise cancellation (ANC) based on neural networks, and for example, recently a new company under the Japanese start-up AI company Qosmo Neutone allows you to use the plugin arbitrarily use AI models to control DSPs;
The operating algorithms are constantly iterating, including three-band dynamic range control technology, dynamic equalizer technology, virtual bass, sound quality improvement technology, surround sound technology, high-definition audio noise reduction technology, etc.;
Combined with the RISC-V instruction set, for example, Haoxin has launched a variety of products in the RISC-V DSP, both fixed-point DSP and floating-point DSP.
There is still a lot of room for localization
The development of DSP in China started late, but the progress has been rapid.
Domestic DSP should start from the beginning of 2000, at that time, the first domestic DSP has been developed, and in the following 20 years, domestic DSP has also experienced reverse design, forward design, compatible substitution to completely independent process.
In 2012, the "Huarui No. 1" domestic DSP project led by the 14th Institute of China Electronics Technology passed the acceptance and carried out large-scale application deployment. In the same year, the "Soul Core No. 1" domestic DSP independently owned by the 38th Institute of China Electronics Technology also completed the test, and its performance can reach the international mainstream level.
Until now, the localization rate of DSP is not high, and the data shows that the corresponding DSP market size in the fields of domestic high-end intelligent equipment manufacturing, communications, and edge computing will reach 16.6 billion yuan in 2022 and is expected to reach 21.9 billion yuan in 2025. The annual demand for DSP chips will be 3.4 billion in 2020, while the domestic production of DSP chips will be 0 in 20209.1 billion, the localization rate is only 26%, in great contrast with the huge market size and demand, with strong market demand potential and urgent independent and controllable, domestic substitution demand.
However, according to a previous article by Leifeng.com, in specific fields, the localization rate is already relatively high, and it is very likely to achieve the goal of 70% in specific fields in 2025. In other fields, the self-sufficiency rate of domestic DSP is less than 10%.
Subdivided into audio DSPs, domestic production is constantly following up:
In August 2022, Fourier Semiconductor completed a new round of financing, and in the official introduction, it is expected to fill the domestic gap in the mass production of automotive-grade audio power amplifier chips;
In August 2023, Xinling Semiconductor completed the closing of Series A financing, which will be used for the testing, certification and mass production of automotive-grade Class D power amplifier chips and the serialization of automotive-grade products.
In October 2023, Jinxin Electronics' high-performance automotive DSP obtained AEC-Q100 reliability certification;
On November 27, 2023, Torch Core Technology announced that Zhuhai Yixin, a wholly-owned subsidiary of Torch Core Technology Co., Ltd., plans to jointly invest in Shanghai Judi Company with Didingrui, Diweisheng and Ruishengtai to help the research and development of DSP chips and technologies of Shanghai Judi Company, which also marks Torch Core Technology's official entry into the field of automotive audio DSP chips;
In November 2023, C*Core Technology successfully developed a DSP chip for high-end cockpit audio processing - CCD5001, and planned a complete series of products, CCD5001 chip products are high-performance DSP chips developed based on HIFI5 architecture core, designed for active noise control, high-end surround sound, intelligent voice interaction and other application scenarios of on-board platforms, the chip is benchmarked against ADI's ADSP-21565 chip, It is developed and produced based on 12nm process.
It can be said that judging from the current actions of domestic manufacturers, the automobile market is also the focus of China, and perhaps in the next year, the automobile market will be an important growth force for audio DSP. With the increasing demand for higher sound quality in automobiles, separate audio DSPs may also become a new demand in the future.