DSP and MCU are moving toward convergence
DSP generally adopts Harvard architecture, super long instruction word architecture, etc., data access and instruction are separated, there are many internal arithmetic units, and there is a special hardware multiply-add structure, so the operation speed is extremely high. Its internal memory (RAM and ROM) is large and expandable, and the external interface is rich, with pipeline operation, which is especially suitable for real-time processing of a large number of digital signals. The MCU data access and instructions are not separated, the operation speed is low, the number of arithmetic units is small, and the internal memory is not large. However, the MCU interface is quite flexible, and integrates FLASH, ADC, DAC, OSC, SRAM, PWM, temperature sensor, watchdog, bus, timer/timer, I/O, serial port and other functional units, so it is very suitable for a variety of control applications.
However, with the increase of system requirements, in some applications, both the system has good control functions and high-speed data processing capability. Therefore, the hybrid processing architecture that combines the advantages of DSP and MCU is undoubtedly a kind of a good solution. DSP and MCU have a common place in practical applications, that is, they are applications for embedded systems, or real-time systems based on the need for large amounts of data processing, or real-time systems that require many control functions. This real-time and versatile also provides a good foundation for the fusion of DSP and MCU. Therefore, the architecture of DSP/MCU convergence is gradually favored by semiconductor manufacturers, and TI, ADI, and Microchip have launched related solutions in an effort to seize the opportunities in this market.
As a global DSP leader, TI has launched a dual-core processing architecture OMAP platform for 2.5G and 3G wireless applications. It integrates ultra-low-power DSP for acceleration applications with ARM925 for control and advanced operating system (OS). )features. The main processor of the OMAP platform is the OMAP1510. The main advantage of its dual-core architecture is that two separate devices are used to perform application processing tasks. That is, ARM925 is used to process control code such as user interfaces, OS and advanced applications, while DSP is used to implement multimedia, voice, security or other functions. The two cores are connected by a dedicated internal communication mechanism. The OMAP platform provides a solution for developing voice applications in portable devices. When used in portable devices, this DSP and MCU combined architecture provides superior performance and power advantages. With optimized underlying software, DSPs can perform signal processing tasks in a low-power manner, extending battery life and enabling smaller product sizes and greatly improving product application performance.
DSP has the advantage of real-time high-speed operation. Its core processing unit has a special structure suitable for digital multiply-add processing, and the "corrected Harvard structure" improves memory management efficiency, and also supports many high-speed peripheral interfaces. The MCU has the characteristics of flexible and efficient control, especially the 16-bit machine that uses the "von-Newman structure". All the memory and peripheral modules are located in the same address space, and the processing capability can far exceed the requirements of the intelligent sensing system. TI's OMPA platform combines the advantages of both. Currently, TI has partnered with a number of major third-party developers who are developing voice technologies such as ASR, TTS, DSR and speaker verification. The company uses TI's OMPA platform solution.
Unlike TI, ADI's DSP/MCU hybrid solution, the embedded processor Blackfin series, uses a single-core architecture. Based on the Micro Signal Architecture (MSA) jointly developed by ADI and Intel, the Blackfin processor combines a 32-bit RISC instruction set with dual 16-bit multiply-accumulate (MAC) signal processing with the ease of use of a general-purpose microcontroller combine it all toghther. The Blackfin processor includes a 10-level RISC MCU/DSP pipeline and a hybrid 16/32-bit instruction set architecture designed for optimal code density. This combination of processing features enables the Blackfin processor to perform well in both signal processing and control processing applications. In many cases, it also eliminates the need to add a separate MCU, simplifying hardware and software design and implementation difficulties. For applications that require both high-performance signal processors and high-efficiency control processors, a Blackfin processor can meet system requirements, reducing development time and cost.
The Blackfin processor architecture also features some of the features of the RISC control processor, including a powerful and flexible hierarchical memory architecture, good code density, and a variety of microcontroller-based peripherals (including 10/100 Ethernet MAC, UARTS, SPI, CAN controller, PWM enabled timer, watchdog timer, real time clock and a seamless synchronous and asynchronous memory controller). These features give designers design flexibility and minimize end system cost. Currently, Blackfin processors are widely used in embedded audio, video and communication applications.
Microchip, which has long been known in the MCU field, has introduced its 16-bit dsPIC digital signal controller (DSC) and first introduced the DSC concept, combining the control advantages of high-performance 16-bit microcontrollers with the high-speed computing of DSPs. Form a tightly coupled single-chip single instruction stream solution for embedded system design.
The dsPIC DSC architecture supports 84 instructions and 10 addressing modes. The dsPIC instruction set consists of a variety of flexible MCU instructions for embedded applications and a set of dedicated DSP operations from a single instruction stream, which can share many CPU resources. The dsPIC DSC core supports various bit operations required for MCU and DSP functions. Bit operations are common in MCUs, but are rarely seen in DSP applications. The dsPIC DSC adds powerful bit manipulation features such as bit testing, bit set and bit shift instructions, and bit lookup operations that recognize the first significant bit in the data word.
Steve Marsh, product marketing engineer for Microchip's Digital Signal Controllers Division, points out that the dsPIC family features some non-digital signal processors (such as barrel shifters or more random access memory), and that's exactly what engineers do. As desired, engineers are more inclined to choose digital signal controllers than microcontrollers for non-digital signal processor applications. The dsPIC family is currently used in AC/DC converters, isolated DC/DC power converters, and other power conversion applications such as embedded power controllers, inverter power supplies, and uninterruptible power supplies (UPS).
MIPS Technologies also introduced the MIPS3224KE, a family of high-performance, low-power cores with built-in DSP extensions that integrates high-efficiency DSP capabilities while significantly reducing overall SoC area, cost and power consumption, and improving signal processing performance. Pete Del Vecchio, MIPS Product Marketing Manager, said the DSP/MCU hybrid processing architecture has performance equal to or higher than the performance of the low-end DSP core. With a single chip or a single core, you can get a significant cost advantage and greatly speed up time to market.
The platform with both DSP and MCU functions was first used in engine control, and then extended to applications such as speech processing and sensor processing, and was used to replace synthetic analog filters with digital filters. Today, this platform is increasingly used in computers, telephone lines or Ethernet and other related fields. In addition, they can be seen everywhere in medical, electrical, air conditioning, uninterruptible power supplies, switched power supplies, semiconductor lighting, and more.
On the one hand, the converged architecture is increasingly used in many areas. On the other hand, the converged platform also faces many problems to be solved:
(1) Power consumption, the platform that combines DSP and MCU has higher performance and higher power consumption than traditional single DSP or MCU. For portable devices that are very sensitive to power consumption, how to further reduce power consumption is the primary problem.
(2) Application environment development, providing users with an easy-to-use development and debugging environment. Zheng Xiaolong pointed out that in terms of hardware, surface-mount QFP and spherical BGA packages began to be widely used, and circuit simulation debugging methods gradually transitioned to boundary-scan interface (JTAG) technology. In terms of software, with the continuous expansion of software scale, it is imperative to use embedded operating system to manage software and hardware resources. The traditional mixed programming mode of C language and assembly language is also introduced, especially object-oriented C++. And the Java language has brought great changes to the traditional development environment. Therefore, it is very important to provide users with an easy-to-use compilation and product development environment.
(3) Cost and design complexity. Embedded systems are becoming more and more complex, so it is becoming more and more important to simplify system design, shorten development cycles, and improve product cost performance.