On November 27, 1990, ARM was founded in Cambridge, England. Now, ARM has gone through 25 years, and the achievements of ARM in the past years are astonishing. As of the third quarter of 2015, the global shipment based on the ARM architecture exceeded 75 billion, and the ARM architecture is applied in 85% of smart mobile devices, especially in the field of smartphones, where 95% of devices use processors designed based on ARM architecture. How did such a semiconductor giant come into being? We provide an in-depth report on ARM!
The History of ARM
Everything Started with 12 People
ARM Holdings, also known as ARM, is headquartered in Cambridge, England. ARM stands for Advanced RISC Machine.
On December 5, 1978, physicist Hermann Hauser and engineer Chris Curry founded the CPU Company (Cambridge Processing Unit) in Cambridge, England, mainly supplying electronic devices to the local market. They were known as the “Apple Computer Company of Britain”.
In 1979, the CPU Company changed its name to Acorn Computers.
Initially, Acorn intended to use Motorola’s 16-bit chip, but found it too slow and expensive. “A machine priced at 500 pounds cannot use a CPU that costs 100 pounds!” They turned to Intel for the design of the 80286 chip but were refused and were forced to develop it themselves.
In desperation, they decided to develop the then-unpopular RISC architecture processor. This choice was made because Acorn was constrained by resources and could not develop a CISC architecture processor, so they could only choose to develop a RISC processor with fewer transistors.
In 1985, Roger Wilson and Steve Furber designed their first generation 32-bit, 6MHz processor, creating a RISC instruction set computer, abbreviated as ARM (Acorn RISC Machine). Later, in the late 1980s, Apple began collaborating with Acorn to develop a new version of the ARM core—ARM2. ARM2 was used in the BBC Archimedes 305. At that time, IBM computers had become mainstream products, and IBM’s sales channels covered the British Isles. Because the BBC Archimedes used a non-mainstream Archimedes operating system, it had almost no market outside the UK except as a successor to the BBC Micro sold to the broadcasting agency and schools.
At this time, Acorn also faced cash flow problems. Against this backdrop, the then IT giant Olivetti acquired Acorn and established an independent subsidiary, Olivetti Research Company, tasked with continuing “a wide range of applied research”. After reaching an agreement with the University of Cambridge, Andy Hopper became the managing director of the company. However, Andy Hopper soon presented a dilemma to his employer, Olivetti.
“We forced them to make a statement. This was a very provocative move. In that situation, I could have been fired,” Andy Hopper later recalled. At that time, he bluntly told Olivetti that he intended to spin off a commercial company from the research company’s lab.
Andy Hopper even proposed the so-called “Andy’s Mutual Wealth Creation Framework”. He explained that whenever a technology might emerge in the lab, the parent company had the right to be the first to obtain it. However, if the parent company does nothing, it would lose that right, and the development work would be spun off into a commercially operated company with researchers and support departments. As compensation, the parent company could hold 20% of the equity in the spun-off company.
The open-minded Olivetti management did not oppose his spin-off proposal, allowing Andy Hopper to establish a model that spun off a new company every two years or so from the lab. On November 27, 1990, with funding from Apple and VLSI Technology (Apple invested 1.5 million pounds, and chip manufacturer VLSI invested 250,000 pounds), Acorn itself invested 1.5 million pounds in intellectual property and 12 engineers, spinning off ARM as an independent subsidiary. So initially, ARM only had 12 people, and due to the poor financial situation of its parent company, the office was just a warehouse.
In the early days of ARM’s establishment, the industry was keen on designing relatively large processors, while ARM, due to limited resources in its design team, had to develop small-scale processors like Acorn before.
ARM’s goal at the time was to quickly launch new products, leading to the design of the ARM610 processor specifically for Apple. This processor was later used in Apple’s Newton PDA. However, this product seemed too advanced for its time, and the market demand was very low, so Newton PDA was discontinued in 1997. Undeniably, the Newton PDA laid a good foundation for Apple’s later successful product, the iPhone.
At the end of 1991, ARM licensed its products to GEC Plessey Semiconductor in the UK. In 1993, ARM licensed its products to Cirrus Logic and Texas Instruments. At that time, TI had achieved great success in the DSP field but was not familiar with the CPU field. ARM collaborated with Nokia and TI to develop a 16-bit Thumb instruction set, creating the ARM/Thumb SoC business model, with ARM7 being the most important processor core, which developed a low-power mode using smaller granules. Subsequently, Samsung also joined the ARM licensing ranks.
In 1995, DEC began developing Strong ARM, and on February 5, 1996, DEC officially released the SA110 processor. This processor gained great fame in the industry. The SA110 processor quickly gained recognition in the industry, and Apple used it in the MessagePAD2000. Also in 1996, the financially troubled Olivetti sold its 14.7% stake in Acorn to Lehman Brothers. In 1996, the ARM8 core was released, doubling performance but could not compete with StrongARM. In 1997, DEC released the second Strong ARM chip SA1100. Intel acquired DEC’s factory in Hudson, and Intel named Strong ARM as X Scale.
In 1997, ARM9 was released, which no longer used the Princeton structure but switched to the Harvard structure, increasing the original 3-stage instruction pipeline to 5 stages, with the highest clock frequency reaching 220MHz. In 1998, the ARM10 core was officially launched, using a 6-stage pipeline structure, improving Cache Memory, optimizing multiplication instructions, and adding floating-point operations. However, due to XScale, ARM10 did not achieve great success, as the XScale processor’s highest operating frequency reached 1.25GHz, and Intel, unable to bear the huge losses, sold the XScale processor business to Marvell. Later, Intel began to promote the ATOM processor.
On April 17, 1998, ARM was listed on both the London Stock Exchange in the UK and the NASDAQ in the United States. Early on, Apple invested $3 million to hold 43% of ARM, but after ARM was listed in the UK and the US in 1998, Apple gradually sold these shares.
Scenes from the company gathering in 2004.
In June 2006, global ARM chip shipments reached 2 billion. By the end of 2007, total shipments of ARM core chips had surpassed 10 billion. Early ARM7 and 8051 used the von Neumann architecture, while current ARM11 or CortexA used the Harvard architecture.
By the end of 2007, ARM had a total of 1,728 employees, held 700 patents (with another 900 pending approval), had 31 global branches, 200 partners, and an annual revenue of £260 million.
In mid-June 2010, Apple expressed interest in acquiring ARM for $8 billion, but was rejected. ARM CEO Warren East stated that ARM as an independent company is more valuable, “The only reason buyers want to acquire is to eliminate competitors.”
On December 13, 2013, ARM announced the acquisition of the well-known lighting engine technology company Geomerics. Geomerics possesses leading technology in the field of lighting and shadow computing. This acquisition will expand ARM’s leading position in the graphics technology industry.
In January 2014, AMD launched the 64-bit ARMv8-A architecture (the first ARM architecture to include a 64-bit instruction set). To date, more than 30 companies are developing 64-bit ARM chips or are about to do so.
In the third quarter of 2015, ARM’s revenue increased by 24% to £243.1 million (approximately $375 million). Third-quarter technology licensing revenue reached $203 million, a year-on-year increase of 35%.
How Does ARM Operate?
ARM has never produced a single chip, but it can be said that the semiconductor industry is basically inseparable from it, and this is inseparable from its business model.
ARM’s Business Model: A Fabless Semiconductor Company Centered on IP Licensing
Unlike Intel, ARM is also a fabless semiconductor, similar to NVIDIA and AMD after selling its wafer factory. However, ARM does not produce any processors; it mainly provides IP licensing to semiconductor partners (such as Qualcomm, Apple, AMD, and Samsung), who use ARM’s architecture, design, and development tools to launch their own processors, which are then supplied to OEM customers (various mobile phone and tablet manufacturers are like this). This is a simple example, and there may also be secondary licensing in between.
ARM’s revenue comes from this cycle, where partners purchasing ARM’s IP licensing need to pay a technology licensing fee (license fee) and will also deduct a certain royalty based on the manufacturer’s processor price (ongoing royalties, which can also be referred to as commission or franchise fees), which may involve various aspects of the chip.
In simple terms, ARM’s business model is “You pay, I authorize”; for processor design and development, it is “All-inclusive teaching” and will provide a series of tools to help customers simplify development.
ARM’s business model is very unique, at least currently, it is very different from the PC market. In the PC market, Intel dominates platform development, and their products usually account for the largest BOM material cost. In the smartphone and tablet market, the main processor cost is mostly within 10% of the total device cost; generally, it is a single-digit ratio. For example, in a $400 device, the SoC processor price is usually $15, accounting for 3.75%. Intel’s theory is that ultra-portable mobile devices will eventually change as chip complexity increases, but so far (or for the foreseeable future), the market still needs different business models.
How Does ARM Operate?
As mentioned earlier, ARM’s main business model is to provide IP licensing, and ARM claims to offer various flexible licensing types. Specifically, there are three licensing methods: POP, processor, and architecture licensing.
Processor licensing refers to licensing partners to use ARM-designed CPU or GPU processors, and the other party cannot change the original design but can use it according to their needs. For example, Samsung’s Exynos 5 Octa uses four Cortex-A7 and four Cortex-A15 processors, which is an example of processor licensing. ARM will provide a series of guidelines to ensure users use their designs, but the final product’s frequency and power consumption, etc., still depend on the manufacturer’s team.
POP (processor optimization pack) licensing is an advanced form of processor licensing. If a partner’s team cannot master ARM processors, ARM can also sell optimized processors to them, allowing users to design and produce performance-guaranteed processors under specific processes. During the Cortex-A8 era, Samsung and Apple’s teams could develop better processors than other companies, but not all companies have such capabilities, so POP licensing is more suitable for those companies that have the intention but lack the ability.
The Cortex-A12 architecture recently released by ARM is a form of POP licensing under GF and TSMC’s 28nm process.
The last licensing method is architecture licensing, where ARM licenses the other party to use its architecture (ARMv7 or ARMv8), and then the other party can design processors according to their needs. This is the licensing method used by Qualcomm’s Krait processors, and Apple’s current “Swift” architecture is also like this. These processors are ISA compatible with ARM’s own designed Cortex-A15 processors, but have Qualcomm’s and Apple’s own implementations.
In this licensing, you will receive some guidance and a series of tests to verify compatibility with ARM ISA. ARM will provide some help, but it cannot help you design and develop your own processors.
ARM currently has over 1,000 licenses and 320 authorized partners, with a quarterly shipment of 2.5 billion licensed chips.
How Does ARM Make Money?
Intel, AMD, and NVIDIA rely on selling processors to make a living, while ARM does not sell any processors. Instead, it mainly relies on technology licensing fees and royalty commissions, which manufacturers must pay. The proportions of these two vary across different processor architectures.
Technology licensing fees are charged based on the complexity of the chip architecture. Older ARM11 is much cheaper than the latest Cortex-A57. The technology licensing fee generally ranges from $1 million to $10 million, but the actual situation may be higher or lower than these two figures.
Royalties are calculated as a percentage, typically ranging from 1% to 2%. If the company’s chip is sold externally, the value is easy to calculate. If sold internally (for example, Samsung’s self-produced products), the royalty percentage is also calculated based on the market price.
Different IP royalty rates
The above shows the royalty rates for different ARM IP licenses, most of which are around 1-2%. The 0.5% rate for POP licensing is not calculated based on chip prices; it is charged to wafer manufacturers, calculated as 0.5% per wafer.
Typically, it takes both parties six months to finalize a contract, and the interval from obtaining technology licensing to the first revenue-generating shipment can be as long as 3-4 years. Depending on market conditions, manufacturers can continue to ship for about 20 years thereafter.
Of the 320 authorized manufacturers, more than half are paying royalties. Most of the other manufacturers are in the stage of signing licensing agreements to shipping products. ARM can add 30-40 new authorized manufacturers each year.
Of the companies that signed licensing agreements, 80% sell the processors they design on the market, while the other 20% either get acquired or fail for other reasons.
50% of ARM’s revenue comes from royalty commissions, 33% from technology licensing fees, and the remainder from software tools and technical support fees.
ARM’s operational situation is good, but overall it is still very small, with a revenue of $913.1 million in 2012. Consider how many ARM designs exist now; ARM should consider raising its royalty rates. However, due to ARM’s unique business model, its gross profit margin is as high as 94%, and its operating profit is around 45%.
Types and Choices of Licensing
Although ARM has already talked about three main licensing types, the licensing provided by ARM is a series of diverse options.
Academic licenses are free, but you cannot sell any designed processors. However, they are great for learning architecture in universities or research institutes. DesignStart is also a low-cost licensing option, but you cannot sell the designed chips for profit.
For those who only need single-use scenarios, ARM offers Single Use licensing, where typical Cortex-A level CPU licensing only requires $1 million and a 2% commission.
Mutil-Use is very meaningful for large companies; although it requires more licensing fees, you can use the CPU licensing in any product within a certain period (for example, 3 years). Unless the license expires, you can use it freely in any product.
Subscription licensing is one of the most interesting licensing methods. Customer companies can spend money to buy a complete set of licenses from ARM. This licensing is suitable for engineering managers who are starting a new chip development project without worrying about the budget. Of course, the cost is higher, with technology licensing fees in the $10 million range.
The top tier is architecture licensing, which has been introduced earlier. Currently, major companies like Marvell, Qualcomm, and Apple have architecture licenses.
Additionally, ARM has to mention the three early partners for collaborative development. Since ARM does not sell any chip products, it needs to ensure that every generation of products has partners releasing processors based on ARM’s latest and best architecture. Therefore, they choose three manufacturers for close cooperation, aiming to expand the market together. They tend to focus on high-end smartphone/tablet SoC processors, but ARM has also found their advantages in industrial applications, digital home, smart TVs, and other markets.
The selected three partner manufacturers can obtain processor architecture information earlier than other licensed manufacturers. They help test and debug and even provide direct feedback to ARM, gaining the benefit of having a better market advantage than other companies.
What Are ARM’s Mainstream Products?
As mentioned earlier, ARM’s main profit comes from IP licensing, so we need to see what products ARM actually has.
In descending order, ARM processors can be roughly ranked as: Cortex-A57 processor, Cortex-A53 processor, Cortex-A15 processor, Cortex-A12 processor, Cortex-A9 processor, Cortex-A8 processor, Cortex-A7 processor, Cortex-A5 processor, ARM11 processor, ARM9 processor, ARM7 processor, and the lower-end products are basically no longer used, so they will not be introduced here.
ARM Processor Architecture Development
● Cortex-A57 and A53 Processors
The Cortex-A53 and Cortex-A57 processors belong to the Cortex-A50 series and are the first to adopt the 64-bit ARMv8 architecture, which is significant. These are the two latest products recently released by ARM.
The Cortex-A57 is ARM’s most advanced and highest-performance application processor, claiming to achieve three times the performance of today’s top smartphones at the same power level; while the Cortex-A53 is the world’s most efficient and smallest 64-bit processor, achieving three times the efficiency of today’s high-end smartphones at the same performance. These two processors can also be integrated into the ARM big.LITTLE (big-little core companion) processor architecture, switching between the two according to computing needs to combine high performance and high power efficiency characteristics, with both processors operating independently.
● Cortex-A15 Processor Architecture Analysis
The ARM Cortex-A15 processor belongs to the Cortex-A series and is based on the ARMv7-A architecture, which is the highest performance and licensable processor in the industry to date.
The Cortex-A15 MPCore processor features an out-of-order superscalar pipeline with tightly coupled low-latency 2-level cache, with a maximum cache size of 4MB. Other improvements in floating-point and NEON media performance enable devices to provide next-generation user experiences and offer high-performance computing for web infrastructure applications. The Cortex-A15 processor can be applied in smartphones, tablets, mobile computing, high-end digital appliances, servers, and wireless infrastructure.
Theoretically, the performance of the Cortex-A15 MPCore processor’s mobile configuration can provide more than five times the performance of current high-end smartphones. In high-end infrastructure applications, the Cortex-A15 can run at speeds of up to 2.5GHz, supporting highly scalable solutions in terms of continuously reducing power consumption, heat dissipation, and cost budgets.
● Cortex-A12 Processor Architecture Analysis
In mid-2013, ARM released the new Cortex-A12 processor, which improves performance by 40% over the Cortex-A9 at the same power consumption while also reducing size by 30%. The Cortex-A12 can also support big.LITTLE technology, further enhancing processor performance when paired with the Cortex-A7 processor.
Cortex-A12 architecture diagram
ARM states that the Cortex-A12 processor will be applied in a large number of smartphones and tablet products in the future, but with a focus on mid-range products. At the same time, ARM also expects that by 2015, the quantity of these mid-range products will far exceed that of flagship smartphones and tablets.
Mid-range devices equipped with the Cortex-A12 processor will also be very distinctive products in the future, as the Cortex-A12 can support virtualization, AMD TrustZone technology, and up to 1TB of onboard storage. This means that smartphones equipped with this processor can be used as BYOD (Bring Your Own Device) devices, meaning they can serve as personal phones while also storing business content.
Mali-V500 architecture diagram
At the same time, the Cortex-A12 is also equipped with the new Mali-T622 graphics chip and Mali-V500 video codec IP solutions, all aimed at energy efficiency. Therefore, targeting the mid-range market, with low power consumption and small size, the Cortex-A12 will inevitably replace the Cortex-A9. It is reported that the Cortex-A12 will be launched in 2014, and we may welcome a change in the mid-range market.
● Cortex-A9 Processor Architecture Analysis
The ARM Cortex-A9 processor belongs to the Cortex-A series and is based on the ARMv7-A architecture. Most of the quad-core processors we see today belong to the Cortex-A9 series.
The Cortex-A9 processor is designed to create the most advanced, efficient, dynamically variable-length, multiple instruction execution superscalar architecture, providing unprecedented high performance and high energy efficiency required by cutting-edge products in a wide range of consumer, networking, enterprise, and mobile applications.
The Cortex-A9 microarchitecture can be used for scalable multi-core processors (Cortex-A9 MPCore multi-core processors) and more traditional processors (Cortex-A9 single-core processors). Scalable multi-core processors and single-core processors support 16, 32, or 64KB 4-way associative L1 cache configurations, with optional L2 cache controllers supporting up to 8MB of L2 cache configuration, providing high flexibility suitable for specific application fields and markets.
● Cortex-A8 Processor Architecture Analysis
The ARM Cortex-A8 processor belongs to the Cortex-A series and is based on the ARMv7-A architecture. It is currently the most common product in single-core mobile phones.
The ARM Cortex-A8 processor is the first product based on the ARMv7 architecture, capable of increasing speed from 600MHz to over 1GHz. The Cortex-A8 processor can meet the power optimization requirements of mobile devices that need to run under 300mW, as well as the performance optimization requirements of consumer applications that require 2000 Dhrystone MIPS.
The Cortex-A8 high-performance processor is now very mature, providing reliable high-performance solutions for high-end feature phones, netbooks, DTVs, printers, and automotive infotainment.
● Cortex-A7 Processor Architecture Analysis
The ARM Cortex-A7 processor belongs to the Cortex-A series and is based on the ARMv7-A architecture. Its characteristic is to provide excellent low power consumption while ensuring performance.
The architecture and function set of the Cortex-A7 processor are identical to those of the Cortex-A15 processor. The difference is that the microarchitecture of the Cortex-A7 processor focuses on providing optimal energy efficiency. Therefore, these two processors can work together in a big.LITTLE configuration, providing the ultimate combination of high performance and ultra-low power consumption. The energy efficiency of a single Cortex-A7 processor is five times that of the ARM Cortex-A8 processor, with a 50% performance increase, while its size is only one-fifth of the latter’s.
As an independent processor, the Cortex-A7 can make entry-level smartphones priced below $100 during 2013-2014 comparable to high-end smartphones priced at $500 in 2010. These entry-level smartphones will redefine connectivity and Internet usage in developing worlds.
● Cortex-A5 Processor Architecture Analysis
The ARM Cortex-A5 processor belongs to the Cortex-A series and is based on the ARMv7-A architecture. It is the most energy-efficient and lowest-cost processor.
The Cortex-A5 processor provides a valuable migration path for existing ARM9 and ARM11 processor designs. It can achieve better performance than ARM1176JZ-S and better efficiency and energy efficiency than ARM926EJ-S. Additionally, the Cortex-A5 processor is fully compatible in terms of instructions and functions with the higher-performance Cortex-A8, Cortex-A9, and Cortex-A15 processors while maintaining backward application compatibility with classic ARM processors (including ARM926EJ-S, ARM1176JZ-S, and ARM7TDMI).
● ARM11 Series Processor Architecture Analysis
The ARM11 series includes ARM11MPCore processors, ARM1176 processors, ARM1156 processors, and ARM1136 processors, which are based on the ARMv6 architecture and target different application fields. The ARM1156 processor is mainly applied in high-reliability and real-time embedded applications, which are not closely related to mobile phones, so it will be omitted here.
The ARM11 MPCore uses a multi-core processor structure, allowing scalability from 1 core to 4 cores, enabling simple system designs with a single macro to integrate up to four times the performance of a single core. The Cortex-A5 processor is a related successor product of the ARM11MPCore.
The ARM1176 processor is mainly applied in smartphones, digital TVs, and e-readers, widely deployed in these fields, providing media and browser functionality and a secure computing environment, achieving speeds of up to 1GHz under low-cost designs.
The ARM1136 processor includes the ARMv6 instruction set with media extensions, Thumb code compression technology, and an optional floating-point coprocessor. The ARM1136 is a mature core widely deployed as an application processor in mobile phones and consumer applications. When adopting a 90G process, the performance can exceed 600MHz, and when adopting a 65nm process with an area of 2 square millimeters, it can reach 1GHz.
● ARM9 Series and ARM7 Series Processor Architecture Analysis
The ARM9 series includes ARM926EJ-S, ARM946E-S, and ARM968E-S processors. Among them, the first two mainly target embedded real-time applications, so we will mainly introduce the ARM926EJ-S.
The ARM926EJ-S is based on the ARMv5TE architecture. As an entry-level processor, it supports various operating systems such as Linux, Windows CE, and Symbian. The ARM926EJ-S processor has been licensed to more than 100 silicon suppliers worldwide and has been successfully deployed in numerous products and applications.
● ARM7 Series Processors
The ARM7 series includes ARM7TDMI-S (ARMv4T architecture) and ARM7EJ-S (ARMv5TEJ architecture), which were first launched in 1994 and are relatively old compared to the above products. Although the ARM7 processor series is still used in some simple 32-bit devices, newer embedded designs are increasingly using the latest ARM processors, which have significantly improved technology over the ARM7 series.
As one of the older series, the ARM7 processor is no longer recommended for use in new products. How old is it? The Apple eMate 300 used a 25MHz ARM7 processor, which is quite old!
Follow us by scanning the QR code below!
On the road of electronics, let’s walk together!