Currently, DSP applications cover communications, medical, multimedia, instrumentation, and artificial intelligence. Data processor requirements continue to present new challenges to processor design teams. For example, in the field of 4G wireless communication, data throughput has reached 100 to 1000 times that of the 3G era; in the field of multimedia processors, new standards emerge every day... With the higher requirements of more new applications, the future In this regard, we specifically interviewed ADI, Atmel, Freescale and Chinese companies in the DSP field, and summarized the following five trends:

First, DSP core instructions have parallel processing capabilities, and SIMD (single instruction multiple data) and VLIW (very long instruction words) will dominate the next generation of high performance processors. Nowadays, almost all DSP manufacturers use the VLIW architecture. This instruction architecture greatly enhances the instruction parallelism of the DSP processor. The SIMD instruction set can greatly improve the data parallel processing capability, and has also been widely used on general-purpose processors. use. For example, CEVA's latest XC323 core includes eight VLIWs, 512-bit SIMD operation, 32 MAC multiply-accumulate operations per cycle, and inherently complex arithmetic operations support. This is CEVA's new architecture for 4G applications, directly targeting high-throughput, high-performance applications.

Second, in some demanding high performance applications, there will be more and more multicore DSP solutions. This is due to the development of semiconductor manufacturing processes, the integration of more and more transistors on a single chip, and the overall multi-core trend of the processor, because of the power consumption, the single core by increasing the frequency It has been difficult to sustain. Relying on multi-core parallel processing to improve performance has become the choice of the entire industry. For example, Texas Instruments' C66x high-performance multi-core DSP family can integrate up to eight C66x cores, each of which can reach 1GHz or 1.25GHz, for a total of 320 GMAC and 160 GLOP fixed-point and floating-point performance, ideal for medical and medical applications. High-end applications such as high-end imaging, test and automation, high-performance computing and core networks.

Third, both fixed-point and floating-point calculations are supported. This trend is consistent with the general direction of the general-purpose processor. The earliest processor is also separated from the integer arithmetic unit and the floating-point arithmetic unit. Now the processor integrates integer and floating-point computing units. From the current product, Kailuo Microelectronics' CLM320C67 series multi-core DSP has integrated fixed-point and floating-point functions. The founder of Qiqi Microelectronics (Beijing) Co., Ltd. said: "In order to ensure high performance, unified fixed-point and floating-point support is easier to program. CHIPLON has already completed this, and we believe that the industry will follow suit." ADI's technology Marketing Manager Zhang Tiehu agrees: "At present, DSP has been moving in the direction of 32-bit, supporting both integer and floating-point arithmetic instruction sets."

Fourth, a single-chip DSP. Single-chip DSP is also known as DSC (digital signal controller), which is also the concept of SOC. It integrates more functions and interfaces through advanced technology to reduce overall board cost, power consumption and size, including various simulations. Digital on-chip peripherals, hardware acceleration, general purpose processing core. Texas Instruments' DaVinci platform is a classic example of a SOC, in addition to Freescale's MC56F82xx series. The MC56F82xx is based on the 56800E core DSC family, which combines the functionality of a microcontroller (MCU) with the processing power of a digital signal processor (DSP) to provide a wide range of peripherals, including high-speed pulse width modulation with NanoEdge layout technology. And two ultra-high speed analog-to-digital converters (ADCs). Charlie Wu, senior systems engineer at Freescale Microcontrollers, said: "This DSC will be used in a wide range of new energy applications, such as digital power, solar power, and battery charging management for hybrid vehicles."

Fifth, DSP + MCU, which is the fusion of microcontrollers. The low cost of the microcontroller is a general-purpose processor that mainly performs intelligent directional control tasks. It can perform intelligent control tasks well, but its shortcoming is that the digital signal processing function is relatively poor, and the DSP can just make up for this functional defect. Many applications require both intelligent control and digital signal processing. Therefore, combining DSP and microprocessor to achieve these two functions with a single chip will greatly accelerate personal communication products, smart phones, and wireless network products. Development to simplify design, reduce PCB size, reduce power consumption and reduce overall system cost. The 32-bit MCU integrated DSP instruction is the development direction. This development is due to cost requirements. Many analog control circuits will be replaced by digital control (software)." Freescale's Charlie Wu explained, "Generally 32-bit." The MCU increases the multiply and accumulate instructions that the DSP has, but only the integer instructions. In this way, the MCU also has some DSP functions, which can perform some filter calculations that are not required for real-time performance. The main architecture of the MCU is a single bus structure. Some Harvard bus structures, that is, a set of program buses plus a set of data buses. In the future, MCUs will evolve to the Harvard bus structure. However, the limitations of this bus structure make it impossible to be as strong as DSP in real-time performance of digital signal processing. ”

The earliest platforms with both DSP and MCU functions were used in the field of engine control, and later extended to applications such as speech processing and sensor processing, replacing the original synthetic analog filters with digital filters. Nowadays, this platform is more and more widely used in computers, telephone equipment and Ethernet-related fields, and it is also in the medical, electrical, air-conditioning, uninterruptible power supply, switching power supply, semiconductor lighting and many other aspects.

The founder and CEO of Chiplon Microelectronics Co., Ltd. said: "With the introduction of a new generation of DSP-enabled MCUs, the difference between entry-level DSPs and ordinary MCUs has become smaller and more blurred." The CLM320C family of 32-bit AVR MCUs with integer fixed-point DSP algorithms, single-cycle multiply and accumulate instructions is a good example.

For a foreseeable period of time, wireless applications will continue to be the driving engine for the programmable DSP market, while the embedded DSP market has more potential. Regardless of the form, DSP technology is still the hub of the emerging IP telephony market. Without DSP technology, there is no access to the Internet and multimedia, and there is no possibility of wireless communication. Therefore, DSP will still be the technology driving force of the entire semiconductor industry.