Source: Yiou.com, CSDN

Organized and published by the Internet of Things Think Tank

Reprint must indicate the source and origin

ARM

The ARM processor is the first RISC microprocessor designed by Acorn Computers for the low-budget market, formerly known as the Acorn RISC Machine. The ARM processor itself is designed as a 32-bit processor but also includes a 16-bit instruction set, generally saving up to 35% compared to equivalent 32-bit code while retaining all the advantages of a 32-bit system.

History of ARM:

On December 5, 1978, physicist Hermann Hauser and engineer Chris Curry founded the CPU company (Cambridge Processing Unit) in Cambridge, England, primarily supplying electronic devices to the local market. In 1979, the CPU company was renamed Acorn Computers.

Initially, Acorn planned to use Motorola’s 16-bit chips but found them too slow and too expensive. “A machine costing £500 cannot use a CPU that costs £100!” They turned to Intel for the design documentation of the 80286 chip but were refused, forcing them to develop their own.

In 1985, Roger Wilson and Steve Furber designed their first generation of 32-bit, 6MHz processors, creating a RISC instruction set computer, abbreviated as ARM (Acorn RISC Machine). This is where the name ARM comes from.

RISC stands for “Reduced Instruction Set Computer,” which supports relatively simple instructions, resulting in low power consumption and low cost, making it particularly suitable for mobile devices. A typical early device using ARM chips was Apple’s Newton PDA.

By the late 1980s, ARM quickly developed into Acorn’s desktop products, forming the foundation of computer education in the UK.

On November 27, 1990, Acorn officially reorganized as ARM Holdings. Apple invested £1.5 million, chip manufacturer VLSI invested £250,000, and Acorn itself contributed £1.5 million in intellectual property and 12 engineers. The company’s office was very humble, just a barn. In the 1990s, ARM’s 32-bit embedded RISC processors expanded globally, occupying a leading position in low-power, low-cost, and high-performance embedded system applications. ARM does not manufacture or sell chips; it only sells chip technology licenses.

MCU

An MCU is essentially a microcontroller, which integrates the CPU, RAM, ROM, timers, and various I/O interfaces onto a single chip, forming a chip-level computer.

Leading manufacturers of MCUs include: Renesas, NXP, New Tang, Microchip, STMicroelectronics, Atmel, Infineon, Texas Instruments, Toshiba, Samsung, Cypress, Analog Devices, Qualcomm, Fujitsu, AMD, Holtek, and many others.

DSP

DSP (Digital Signal Processing) refers to the theory and technology of processing signals through numerical computation. Additionally, DSP is short for Digital Signal Processor, a chip that integrates a dedicated computer, roughly the size of a coin.

FPGA

FPGA (Field-Programmable Gate Array) is a product developed based on programmable devices such as PAL, GAL, and CPLD. It emerged as a semi-custom circuit in the ASIC field, solving the shortcomings of custom circuits and overcoming the limitations of programmable devices.

Leading manufacturers of FPGAs include: Altera (acquired by Intel), Xilinx, Actel, Lattice, Atmel, QuickLogic, Microsemi, Cypress, TI, and many others.

SoC

The definition of SoC varies widely; it is rich in connotation and application scope, making it difficult to provide an exact definition. Generally speaking, SoC refers to a system-on-chip, indicating it is a product, an integrated circuit with a specific purpose that contains a complete system and all embedded software content. It is also a technology used to implement the entire process from defining system functions to software/hardware partitioning and design completion.

Comparison of ARM, MCU, DSP, FPGA, and SoC

1. Architecture

ARM: Uses a 32-bit RISC processor architecture. From ARM9 onwards, ARM adopted the Harvard architecture, which separates instructions and data into their own independent memory structures, significantly improving the processing capability of the processor. ARM often uses pipelining technology to shorten program execution time, allowing instructions to flow through multiple pipelines, thus enhancing processor efficiency and throughput. Today, ARM7 uses a typical three-stage pipeline, ARM9 uses a five-stage pipeline, and ARM11 uses a seven-stage pipeline, with ARMCortex-A9 supporting up to four cores, marking the first support for multi-core technology in the ARM series of processors.

MCU: Most are based on the von Neumann architecture, which clearly defines the four basic parts necessary for embedded systems: a central processing unit core, program memory (ROM or flash), data memory (RAM), one or more timers/counters, and input/output ports for communication with peripheral devices—all integrated onto a single integrated circuit chip. Early MCUs used CISC instruction sets, later replaced by RISC. MCUs cover bus widths of 4-bit, 8-bit, 16-bit, and 32-bit, with widespread applications.

DSP: Also known as digital signal processors, these are microprocessors specifically designed for real-time digital signal processing. Structurally, they adopt Harvard architecture and also utilize pipelining technology. Additionally, DSP can operate as a direct memory access device in host environments and supports data acquisition from analog-to-digital converters (ADC), ultimately outputting data converted to analog signals by digital-to-analog converters (DAC), supporting certain parallel processing capabilities.

FPGA: FPGA stands for Field Programmable Gate Array, which is a product further developed from programmable devices such as PAL, GAL, and PLD, and is the highest integration of ASICs. FPGA uses a new concept called Logic Cell Array (LCA), which includes Configurable Logic Blocks (CLB), Input/Output Blocks (IOB), and interconnects. Users can reconfigure the logic modules and I/O modules inside the FPGA to achieve their logic. It also features static reprogrammability and dynamic in-system reconfiguration, allowing hardware functions to be modified through programming like software. FPGA’s main distinction from DSP, ARM, and MCU is its parallel processing capability, which greatly speeds up complex computations.

SOC: A system-on-chip integrates a computer or other electronic systems onto a single chip. SoCs can handle digital, analog, mixed signals, and even higher frequency signals. SoCs are often used in embedded systems and have a large integration scale, generally reaching millions to tens of millions of gates. SoCs are relatively flexible, integrating ARM architecture processors with specialized peripheral chips to form a system. Some ARM processors like Hisi-3507 and hisi3516 are SoC systems, especially application processors that integrate many peripheral devices to provide robust support for executing more complex tasks and applications.

Xilinx’s ZYNQ architecture chip

2. Power Consumption

ARM: ARM’s significant success in the mobile market is mainly due to its low power consumption. It is well known that electronic products in the mobile market are very sensitive to processor power consumption. In the past, processor power consumption on PC platforms ranged from several dozen watts to over a hundred watts, which is unimaginable for mobile platforms. ARM’s power consumption at 1GHz is only a few hundred mW, making it suitable for mobile electronic products.

DSP: According to a set of data from the non-network sector, DSP and FPGA each hold half the market share in digital signal processing. One advantage of DSP over FPGA is its relatively low power consumption. DSP manufacturers increase the processor’s clock speed while striving to reduce power consumption to maintain market share because FPGA seems to hold an advantage in high-performance digital processing markets. Looking purely at the DSP field, TI’s DSP processors perform best in terms of power consumption and performance.

MCU: MCUs have been around the longest, with various manufacturers having their own architectures and instruction sets. In terms of low power consumption, TI’s MSP430 MCU performs relatively well.

FPGA: Due to its internal structure, FPGAs have relatively high power consumption and heat generation, which is a downside. However, this is unavoidable in supporting high-performance concurrent computing digital circuits, and the logic gates mainly use standard aspect ratios, resulting in power consumption that cannot compete with ASICs and other dedicated processors.

SOC: Due to the flexibility of SOC, it integrates multiple devices into a very small chip to form a system, giving SOC systems an advantage in power consumption compared to systems composed of MCUs and other processors. Additionally, SOC chips can optimize system power consumption systematically considering factors like processes and circuit designs at the layout level, resulting in lower power consumption and smaller footprint than systems built with current peripheral PCBs.

3. Speed

As market demands increase, ARM manufacturers are optimizing to raise clock speeds and enhance performance, from the initial 100MHz to an astonishing 2.3GHz, with ARM clock speeds advancing rapidly.

Currently, the fastest DSP can reach a clock speed of 1.2GHz. However, clock speed alone does not determine performance superiority over ARM; DSPs can complete a multiplication and an addition in one clock cycle, a capability that typical ARMs do not possess, giving DSPs a clear advantage in computational fields. TI has combined the strengths of both ARM and DSP to produce the Da Vinci heterogeneous chip, which falls into the SOC category.

MCUs, as low-end application processors, have clock speeds ranging from a few MHz to several tens of MHz.

FPGAs can achieve clock speeds of several GHz, even exceeding 10GHz, though they come with a high cost. Comparing clock speeds between FPGAs and ARM, DSPs, etc., is not very meaningful, as the parallel computing capabilities of FPGAs far exceed those of typical general-purpose processors using serial computing by dozens of times. For instance, implementing the same filtering algorithm on a 100MHz FPGA is significantly faster than on a 1GHz ARM.

4. Applications and Markets

ARM processors are now mainly divided into three series: A series, R series, and M series, with the A series focusing on consumer electronics applications and being widely used.

The R series processors are mainly aimed at applications requiring high real-time performance, such as aerospace and automotive electronics, featuring high reliability, high availability, fault tolerance, and real-time response.

The M series processors target lower-end applications, originally meant to replace existing MCUs on the market.

DSPs are primarily aimed at applications requiring high computational power, such as video image processing, intelligent robotics, digital wireless, broadband access, digital audio, high-resolution imaging, and digital motor control.

MCUs are the most widely used due to their cost-effectiveness, allowing them to thrive in many applications with lower computational demands. In the coming years, the key growth drivers for the MCU market will come from green energy, smart electronic devices, smart grids, and upgrades in electronic products like automotive electronics.

SoCs are also widely used, mainly because the architectures of mainstream ARM chips are a type of SoC architecture. SoC is a broad concept, and currently, many ARM and DSPs are beginning to adopt SoC methods to integrate multiple devices into processors to form complex systems.

5. Development Costs

ARM primarily runs on operating systems like LINUX, ANDROID, and WINCE. In terms of development difficulty, it is relatively more challenging to learn than MCUs and DSPs, requiring developers to have a deep understanding of operating systems. In terms of cost, ARM’s single-chip cost is higher than that of MCUs, mainly applied in more complex systems.

MCUs are the easiest to get started with, quick to learn, and have lower development difficulty, making them widely used in low-end markets due to their low cost.

DSPs are relatively easy to learn, but their single-chip costs are higher, primarily used in applications requiring high computational power. DSPs can also run operating systems, making them suitable for multitasking applications.

FPGA development is relatively challenging and has a longer development cycle, in addition to having a high single-chip cost.

Example: SOBEL operator (horizontal edge)

Typically, performing such an operation requires 9 multiplications and 8 additions. This computation is quite easy for FPGAs and DSPs, but for ARMs and MCUs, their weak parallel capabilities make them struggle when processing larger images, such as 1280P.

However, such operations can be easily optimized. For example, the pixel points at positions 1 and -1 can be completed with a single addition, and similarly for the last row. The corresponding pixel points of 2 and -2 in the middle row can also be added once and then shifted, reducing the computation from 9 multiplications and 8 additions to three additions and one shift (the shift operation can generally be completed in a single clock cycle on most processors).