The past of chip architecture.

Recently, chip design company MIPS announced that it has hired two former executives from the well-known RISC-V technology company SiFive. Drew Barbier has been appointed as the company's Vice President of Product, and Brad Burgess has been appointed as the company's Chief Architect to promote the development of its eVocore series core IP based on the RISC-V instruction set architecture.

The two experts have very rich industry experience in semiconductor IP and the RISC-V field. In light of this, MIPS CEO Sameer Wasson said: "I am very confident in their ability to help the company drive IP innovation and penetrate new markets (RISC-V)."

In short, MIPS has poached SiFive's talent, aiming to strengthen RISC-V research and development.

Friends who are familiar with processor architecture should know that MIPS is the founder of the MIPS instruction set architecture. After 40 years of industrial changes and industry turmoil, MIPS has abandoned the MIPS architecture and transformed into a RISC-V CPU IP design company.

As a well-known promoter of reduced instruction set computing (RISC), the establishment of MIPS is even six years earlier than its competitor ARM, let alone the latercomer RISC-V. However, seeing ARM and RISC-V thriving in various fields and continuously increasing their application scale, the decline of MIPS is somewhat puzzling.

Advertisement

Taking this opportunity, let's review the brilliant and low points in the 40-year development process of the MIPS architecture, as well as its subsequent transfer of hands, gradual decline, and finally being forced to withdraw from the architectural competition and join RISC-V's "stormy journey."

Moreover, in the long history of chip industry development, which processor architectures have been sought after and favored by the market, and ultimately disappeared in the rolling wheels of history.

MIPS "once popular"

Looking back at the development of the industry, we can see that the microprocessors at the core of various modern electrical appliances originated from three plans of three companies, Intel, Texas Instruments, and Garrett AiResearch's industrial department in the last century. The three pioneers of the microprocessors launched were the Intel 4004, TMS 1000, and CADC.

Since then, the microprocessor revolution has started.In the early days of the microprocessor, it was essentially different manufacturers producing chips with different architectures. Although later under the "coercion" of IBM, Intel licensed the X86 architecture to several other manufacturers to produce processors, the mainstream at the time was that processor manufacturers independently developed their own architectures and designs, and even produced processors themselves.

This situation continued until around 1980, until the advent of Reduced Instruction Set Computer (RISC).

In 1980, IBM's 801 project may have been the first system designed using the concept of a reduced instruction set.

Unlike the complex instruction set of Intel X86, it designed many features to make code writing more convenient, but these complex features require several instruction cycles to be implemented and are often not used by the running program. In addition, the speed difference between the processor and the main memory is also becoming increasingly large.

Under the influence of these factors, a series of new technologies emerged that allowed the processor's instructions to be executed in a pipeline, while reducing the number of times the processor accessed memory. RISC has streamlined the number of instructions and addressing methods, making it easier to implement, better parallel execution of instructions, and higher compiler efficiency.

The emergence of this relatively new instruction set has promoted the birth of MIPS.

After RISC became popular, Professor Hennessy of Stanford University led the team to develop the first MIPS architecture processor, and this project was the predecessor of MIPS.

In 1984, Professor Hennessy left Stanford University to establish MIPS. Since then, one of the leaders of the reduced instruction set architecture, the MIPS architecture, has stepped onto the historical stage.

In the second year of its establishment, MIPS launched the first processor design R2000. Since then, MIPS has launched several generations of very successful commercial products, such as the R3000 with sales exceeding one million, and the R3000A with sales exceeding one hundred million.

It is worth noting that MIPS even launched a 64-bit design R4000 as early as 1991. It is known that its competitor ARM did not widely promote the 64-bit processor design until 2012. This shows its foresight in design.In 1992, SGI, which had high hopes for MIPS, acquired it and supported MIPS to continue to launch several generations of R-series processors in the 1990s. This was the first change of hands for MIPS, and after that, MIPS also entered a pattern of instability.

Around the year 2000, the release of the MIPS32 and MIPS64 instruction sets marked that 32-bit and 64-bit MIPS CPUs were used more and more widely.

MIPS processors were a major hotspot in the RISC CPU design in the mid-1980s. Semiconductor companies such as Pace MIPS, IDT, and Toshiba all used MIPS designs to manufacture chips, and the chips they produced were also adopted by terminal devices such as Sony and Nintendo's game consoles, Cisco's routers, and SGI supercomputers, especially in the home router market, which was almost absolutely monopolized by MIPS.

Compared with X86, the licensing cost of MIPS is lower, which is also adopted by most chip manufacturers other than Intel. In addition to the lower cost, there are two more reasons why MIPS has been sought after for nearly 40 years:

MIPS's architectural licensing does not restrict any changes to the MIPS architecture.

MIPS belongs to the reduced instruction set, and compared with the complex instruction set computing structure (CISC) adopted by Intel, RISC has the advantages of simpler design, shorter design cycle, and can apply more advanced technologies to develop faster next-generation processors.

Although it has not surpassed X86 in mainstream PC/server applications, it has also achieved hundreds of millions of CPU sales through home game consoles such as PlayStation, PlayStation 2, and Nintendo 64.

MIPS "declining day by day"

However, water can carry a boat and also overturn it.Due to cost issues and the limitations and drawbacks of the licensing model, MIPS was unable to compete with major companies in the market and began to fall into a research and development predicament.

MIPS was initially positioned as a product to compete with Intel's X86 architecture, and therefore its products have been known for high performance since their inception, making significant achievements in the markets for high-definition set-top boxes, routers, and the like.

Although the products have very good performance, the lack of business acumen led to the commercialization process of MIPS falling far behind.

In addition, compared to its competitor ARM, which has been targeting the embedded low-power field since its inception and has been working hard in this field for more than a decade, it finally ushered in its own era in the 21st century. At that time, MIPS, which only focused on the mid-to-high-end market, did not have much advantage in power consumption, which greatly limited its development.

And the slow response of MIPS once again hindered their transformation.

When ARM, in conjunction with companies such as Qualcomm, Apple, and MediaTek, was targeting the smartphone market to develop mobile processor chips, MIPS was still intoxicated in the niche markets for high-definition set-top boxes, printers, and other products, lacking the motivation to revolutionize itself, and as a result, was revolutionized by ARM.

Later, the big explosion in the smartphone market caused it to miss the era of the mobile market, and it officially sounded the "death knell" for MIPS.

In summary, the sluggishness of MIPS led them to lose the most critical decade.

In addition, industry insiders also said that the licensing model is also a major factor in the failure of MIPS.

Both MIPS and ARM have two licensing models: IP licensing and architecture licensing, but the ideas of the two sides are very different.ARM's licensing approach is highly flexible (mostly IP licensing, with very few architectural licenses), and it has a price advantage, which has attracted the attention of companies such as TI and LSI. At the same time, the relatively low price has attracted more people to play with ARM chips, improving ARM's toolchain and ecosystem, and laying a solid foundation for ARM's later explosion.

In contrast, MIPS mainly adopts architectural licensing (IP licensing is very expensive). Although this is more open, it leads to customers fighting on their own and designing their own MIPS cores, adding instructions, and releasing development tools, resulting in severe fragmentation and difficulty in forming a unified ecosystem before fully occupying the market.

At the same time, this also loses the advantages of IP licensing, such as fast launch speed and high software compatibility. The backwardness of hardware leads to the backwardness of the software platform, creating a vicious cycle that accelerates the decline of MIPS.

In the following years, MIPS gradually declined after several changes of hands by Imagination Technologies and Tallwood Venture Capital.

At the end of 2012, MIPS was divided and acquired by ARM and Imagination.

For ARM, MIPS patents are quite valuable, especially those related to 64-bit and multithreading. Because ARM's 64-bit architecture has a 70-80% similarity with MIPS 64-bit, if it does not participate in the acquisition at this time, it may fall into a long-term patent lawsuit with MIPS in the future. ARM is very happy to solve this potential hidden danger with only 350 million US dollars. Therefore, ARM acquired nearly 500 patents.

Imagination acquired MIPS to strengthen its own CPU business and valued MIPS's strong product set, as well as support for the Android architecture and authorization in China. Therefore, Imagination acquired the MIPS company entity and 82 core patents related to the MIPS processor core architecture. At the same time, as the largest shareholder of Imagination, Intel's acquisition of MIPS can also restrict the development of ARM from the side.

After acquiring MIPS, Imagination planned to imitate ARM's operation model by combining its own PowerVR product series. Although MIPS still lags behind ARM in the IoT market share, it has found many opportunities in the automotive, data center, and other markets, such as the MIPS CPU used by Mobileye.

In those years, Imagination accelerated the pace of launching MIPS processor cores, especially the Warrior series that supports hardware virtualization, multithreading, and SIMD.

However, the good times did not last long. In April 2017, Imagination announced the termination of cooperation with its largest customer, Apple, and its stock price shrank instantly, falling into a difficult situation. At that time, Imagination, who was "hard to save itself," was looking for a buyer for itself, and also sold MIPS to Tallwood Venture Capital for 65 million US dollars separately, until it was acquired by Wave Computing in 2018.After being acquired by Wave Computing, MIPS did not take a better path.

At this time, in addition to the already mature X86, almost every ISA is planning to leverage the power of the open-source community to develop its ecosystem, and MIPS is no exception.

At the end of 2018, Wave Computing announced the launch of the MIPS Open Initiative, which will be open-sourced in the first quarter of 2019 when the latest Core R6 is released. Participants in the project can try the 32-bit and 64-bit MIPS ISA for free without any licensing and patent fees.

This move aims to accelerate the popularization of the MIPS instruction set architecture, hoping to help the gradually marginalized MIPS instruction set architecture get back on track. At that time, the open-source RISC-V architecture also began to gradually receive market attention.

MIPS's open-source move was considered at the time to both show off to ARM and declare war on RISC-V, and to prepare for the upcoming IoT era.

But choosing open-source means that it is necessary to accumulate strength and accompany time, such as Linux, RT-Thread, TiDB and other successful open-source projects, all have a long-term accumulation of community popularity and technical precipitation, and wanting to achieve immediate results and eager for quick success in open-source can only backfire.

However, this open-source project did not last a year and ended with the adjustment of the company's CEO. At the end of 2019, the short-term open-source of MIPS officially ended, which further hit the enthusiasm of developers for MIPS.

In this process, the entanglement between MIPS and Loongson, Shanghai Xinlianxin, seems to have also become an episode in the process of the decline of the MIPS architecture, adding a point for people to speculate and judge.

Not long after the open-source project was announced to end, Wave Computing chose to apply for bankruptcy protection and reorganize assets.

In 2021, Wave Computing completed the bankruptcy reorganization and changed its name to MIPS, but the "reborn" MIPS announced that it would no longer design MIPS processors in the future, but turned to the development of RISC-V processors.Can MIPS Write a New Chapter?

Once a contender alongside X86 and ARM, the MIPS architecture has undergone several transformations and has now reached a final chapter in its destiny, officially joining the ranks of the RISC-V, which is also a reduced instruction set.

According to the official MIPS website, it is developing RISC-V designs based on the innovation of the MIPS ISA. Since there are many similarities between the RISC-V ISA and previous generations of the MIPS ISA, it can bring decades of development experience to its new eVocore product series.

At present, MIPS has launched two RISC-V CPU IPs: the eVocore P8700 for high-performance applications and the eVocore I8500 for high-energy efficiency applications.

MIPS has expressed its intention to enter a broader market with its new RISC-V IP, including applications in automotive, machine learning, 5G, data centers, storage, and high-performance embedded systems.

It is reported that Mobileye is already a customer of the P8700. Not long ago, MIPS also signed an agreement with an undisclosed customer in the data center market. It is said that the customer is "one of the three hyperscale computing companies."

Although it is still unclear how many manufacturers will adopt MIPS's RISC-V IP core, the fact that MIPS has recruited two top talents from SiFive during its business restructuring also reflects MIPS's continuous efforts to fully commit to RISC-V.

MIPS once stated about its open source behavior, "If this had happened two or three years ago, then RISC-V would never have been born."

But if following a false premise can lead to any proposition, the word "if" can lead to any result. However, there are not so many "ifs" in the world. Looking at the current situation, RISC-V is developing smoothly, while MIPS has already changed its course. Even Imagination, which acquired MIPS more than ten years ago, has now stepped into the RISC-V camp.Today, under the leadership of the new CEO Sameer Wasson, the primary task is to tell the story of MIPS' rebirth, how to achieve differentiation within the RISC-V community, and where it plans to focus its business.

This time, if MIPS can grasp the lifeline of RISC-V, it may still have a place in the future.

However, as you know, there are not so many "ifs" in the world.

Processor architecture, leading the trend for decades

The changes that have occurred in MIPS foretell another processor architecture bowing to fate. Similarly, as one of the most historically significant processor architectures, MIPS has witnessed many other processor architectures undergoing changes - watching X86 dominate the PC market, ARM rise in the mobile market, and RISC-V, as a rising star, being favored and pursued by the industry; but at the same time, MIPS has also witnessed some architectures fading away.

Previously, the semiconductor industry observation once mentioned in the article "The History of Processor Architecture Extinction" that processor architectures born around the 1980s, including the well-known X86, MIPS, and ARM, also include SPARC, DEC Alpha, PA-RISC, and other products that emerged at the same time. Most of these architectures originated from the RISC system, including IBM's Power architecture, which is also a member of the RISC system.

The arrival of these architectures has built a splendid era of processor architecture and exerted a certain pressure on Intel at that time, fearing to shake its X86 position.

Today's RISC ISA, in addition to the thriving ARM and RISC-V, may only be able to make a little splash with IBM's Power.

PowerPC architecture

PowerPC is a microprocessor architecture developed by the AIM alliance composed of Apple, IBM, and Motorola in 1991.The PowerPC architecture is a processor architecture based on Reduced Instruction Set Computing (RISC), designed with a focus on performance, low power consumption, and scalability. It employs techniques such as pipeline execution, superscalar execution, and out-of-order execution to enhance the efficiency of instruction execution. The PowerPC architecture also supports multi-level caching and high-speed buses to accelerate data transfer and access speeds, aiming to provide high performance and scalability for personal computers, workstations, and servers.

At the same time, the PowerPC architecture boasts excellent cross-platform compatibility because its instruction set is unified. This means that programs compiled on different PowerPC processors can run on various PowerPC systems without recompilation, bringing convenience to software developers and users.

Thanks to these excellent characteristics, the PowerPC architecture later became the main processor architecture for Apple Macintosh computers, being used from 1994 until 2006, after which Apple switched to processors based on the Intel X86 architecture.

For a period, the PowerPC achieved tremendous success. The Motorola 68000 series chips were the core of Apple PCs and many types and millions of embedded controllers. In addition to Apple, Sony's PlayStation 3, Nintendo's Wii, Wii U, GameCube, Microsoft's Xbox 360, and 3DO M2 all used PowerPC processors.

In fact, when Apple abandoned PowerPC, in terms of quantity, PowerPC did not lose a large number of customers, but it lost its most prestigious customer. Then the game consoles also abandoned PowerPC, and typical embedded systems also abandoned it.

Looking back at the 1990s and early 21st century, licensing processor IP was all the rage, which was the most reliable way to achieve widespread adoption. In fact, from the beginning, IBM followed the licensing path of PowerPC, emulating architectures such as ARM, MIPS, and SPARC. However, IBM's terms were too harsh, and PowerPC licenses were much more expensive than MIPS or ARM licenses.

The failure of PowerPC to become widespread in the market was largely due to insufficient openness and IBM's high licensing fees. After the era of smartphones, PowerPC gradually became marginalized due to cost issues.

Although for various reasons, the Power architecture is no longer so dazzling in everyone's sight, its history of competing with X86 cannot be erased. Currently, manufacturers including Guoxin Technology are still using this architecture to design chips.

SPARC Architecture

SPARC (Scalable Processor Architecture) is a Reduced Instruction Set Computing (RISC) instruction set architecture (ISA) originally developed by Sun Microsystems. Its design is largely influenced by early RISC designs, which are minimalist, including as few functions or operation codes as possible, aiming to execute instructions at a rate of almost one instruction per clock cycle.SPARC was first developed in 1986 and released in 1987, and it is one of the most successful early commercial RISC systems. After launching its first SPARC processor product, SPARC quickly captured the market and helped the company break through the $1 billion revenue threshold.

By 1989, processors using the SPARC architecture began to be applied to high-performance workstations and servers. The openness of the architecture and the characteristics of the RISC system quickly made it an internationally popular architecture.

At the same time, to expand the influence of SPARC and make further optimizations, the "SPARC International" organization was established, including Motorola, Toshiba, Fujitsu, and Aeroflex Gaisle. In 1995, with the launch of UltraSPARC I, Sun's leading position in the high-end microprocessor market was further consolidated.

However, after entering the new century, the rise of new applications made the PC no longer the only battlefield for processor architecture competition. The ARM architecture, with its simple instruction set, entered the mobile market and gradually became the dominant player in this market.

At this time, Intel's X86 had established a strong ecosystem over the past few decades. Therefore, without grasping the arrival of new applications and under the pressure of Intel's capture of the vast majority of the market, many processor architectures, including SPARC, gradually began to decline.

With the rise of the Internet, not only did it promote the development of the mobile market, but it also promoted the demand for servers. However, it was in this market that SPARC fell behind. According to Gartner data, from 2002, Sun Microsystems' revenue share has been declining, and by 2007, it was officially overtaken by IBM. The RISC+UNIX server market was also gradually replaced by Intel's X86+Linux/Windows.

Due to the failure in the server market, SPARC began to decline.

In 2010, Oracle acquired SUN for $7.4 billion, and SPARC also belonged to Oracle. After 2017, Oracle was exposed to layoffs in the SPARC department, and gradually, Oracle also gave up the development of SPARC.

Alpha architecture

At that time, while many RISC architectures were emerging, DEC was also influenced by this RISC trend.Thus, between 1982 and 1985, DEC divided the RISC into several parts for separate research. In 1985, one of the leaders in charge of RISC research at DEC proposed the "Collaborative RISC Project," which merged the previous projects into one and renamed it PRISM.

During this period, in order to further smoothly introduce the new architecture to the market, DEC began to consider whether it was necessary to produce a new computer architecture to replace its existing VAX product line, and finally ended the PRISM project in 1988.

At this time, processor architecture began to develop towards 64 bits. The development towards 64 bits was also seen by many manufacturers at that time as a turning point that could rewrite the market pattern. Therefore, DEC began to consider using RISC-like design concepts to design a new generation of VAX CPUs to improve speed while expanding to a complete 64-bit architecture.

Thus, in 1992, the Alpha architecture was born.

Alpha, as a microprocessor architecture designed specifically for high-end desktops, workstations, and servers, also made them one of the first companies to achieve a 64-bit architecture.

Looking back now, Alpha's contribution to the industry may not lie in how many products have adopted this architecture, but rather in its theoretical provision of a new way of thinking for the development of the industry.

Some people believe that Alpha and MIPS, two RISC architectures, considered 64 bits earlier and introduced many advanced micro-architecture design concepts, which influenced the future development of Intel in micro-architecture and hyper-threading, and some shadows of Alpha architecture can be seen in the micro-architecture design of Intel processors.

SUN, as a representative of open-source architecture, is regrettable for its decline, but the demise of DEC Alpha was due to its overly closed ecosystem.

According to data, DEC produced all accessories and peripherals related to Alpha processors by itself, but the motherboard developed for desktop computers did not support SMP, and almost all companies using Alpha processors at the time would use multi-processor systems, so the desktop models launched by DEC were not competitive.

In addition, Alpha processors have always not supported free open-source operating systems, which has also become another reason for their defeat.In 1998, due to financial reasons, DEC sold the Alpha architecture along with most of its content to Compaq. However, Compaq, which was already an Intel customer, decided to phase out Alpha in favor of the upcoming Itanium architecture and sold all Alpha intellectual property to Intel in 2001.

Since then, the story of Alpha has come to an end.

PA-RISC Architecture

In addition to this, the PA-RISC architecture introduced by Hewlett-Packard also could not escape the fate of being eliminated.

PA-RISC was also a member of the mid-1980s, first appearing on February 26, 1986, and was applied to the HP 3000 930 series and HP 9000 840 mode processors. Later, this architecture was replaced by the Itanium architecture jointly developed by Hewlett-Packard and Intel.

However, Intel's Itanium did not become the ultimate winner either. This 64-bit architecture, named in 1999, has not been updated since 2017. In 2019, Intel announced that the Itanium 9700 processor would begin to be phased out, with the last shipment in July 2021, and HPE's service area will be supported until 2025.

With the discontinuation of pure 64-bit Itanium processors, it also means that the Itanium architecture has faded into history.

It turns out that in the process of market development, no processor architecture can always meet the needs of the market. However, from the current results, RISC-V has become the overlord of other RISC designs and the forgotten architectures along the way, becoming their final destination.

In conclusion, looking back at history from the current perspective, in terms of processor design and energy efficiency ratio, if there is a most classic RISC processor architecture, it must be MIPS. Even its competitors have to admit its elegance. MIPS is regarded as a model in processor textbooks, and its presence can be seen in many other processors.The historical stories of MIPS, including other architectures such as SPARC, Alpha, and PA-RISC, have precisely demonstrated the "uncertainty" of the semiconductor market. An ISA that cannot establish a solid foothold and keep up with the industry's development needs will quickly be eliminated by the market.

Looking at the current processor architecture market, X86 still stands firm, but the players competing against it have changed. In this round, the competitors are the ARM and RISC-V camps. Looking at the development of the latter two, they have been actively expanding into the PC market in recent years. At the same time, driven by the era of big data, they also have intentions to develop in the direction of the server market.

But for this market, Intel has already established a strong advantage in this field. In addition to the products, the ecosystem is also a powerful weapon for it to defend its territory.

However, emerging markets are always developing, industry needs are endless, and the business world is changing rapidly. This elimination competition of processor architecture seems to have not ended.

No one should think they are stable.

Post a comment