Microsoft to Develop ARM Server Device
09-01-2021 | By Robin Mitchell
In recent news, Microsoft has announced that it will be developing its own ARM-based processors for servers, and potentially Microsoft Surface devices. How has Intel fallen behind the integrated circuit race, why are more companies going fabless, and what advantages does designing custom hardware provide?
Microsoft Announces its ARM Plans
Recently, Microsoft announced that it would be designing its own semiconductors for use in their servers that run their cloud platform Azure. Currently, Azure utilises Intel processors, but the move would see ARM architecture, which is a major move when considering that Intel utilises CISC (Complex Instruction Set). In contrast, ARM utilises RISC (Reduced Instruction Set). The same announcement also mentions that Microsoft may consider using the same technology to drive its Surface range of devices.
Most Microsoft software operates on x86/x64 processors, but a few editions work on the ARM instruction set (these are mainly aimed at the IoT industry). Furthermore, Microsoft has already worked with Qualcomm to produce mobile ARM-based devices for Surface Laptop 3 and Surface Pro X.
If Microsoft goes ahead with the move to switch to ARM and develop its own hardware, it will be following in the footsteps of other companies such as Apple who have already developed their own ARM-based processors for use in computing applications. Why would a company want to design their own semiconductor hardware, and how have companies such as Intel fallen behind in the processor race?
Custom Hardware
Go back 40 years, and you would find that companies which designed semiconductors also manufactured them. For example, an Intel processor would be designed and made by Intel, while Texas Instruments would also make a 4001 quad NOR IC designed by Texas Instruments.
However, the introduction of uncommitted logic arrays (ULA) allowed companies to design custom silicon designs (with limitations). While initially expensive, these designs would allow for a circuit to be reduced in complexity significantly, and provide unique functions that would otherwise be too difficult to build using off-the-shelf logic chips. For example, many 1980’s computers would use ULA’s to control cassette tapes for storing data and displaying graphics on a TV.
The popularity of designing silicon devices but not manufacturing led to the first fabless companies (Xilinx was the first). Such a company specialises in designing silicon products but does not manufacturer them. This separation of design and manufacturer presented many opportunities and advantages that have continued until this day.
Having access to custom silicon allows for designers to create a circuit perfectly tailored to their devices needs. Circuit functions that are not needed can be removed, thereby reducing power consumption. The leftover space from removing such circuits can either be dedicated to increasing the capabilities of important functions or reducing the design's overall size.
Furthermore, the use of custom silicon allows designers to choose from a wide range of different off-the-shelf designs that can be glued into the circuit in a near-identical fashion to 1970’s glue logic. For example, a designer can choose to integrate an ARM processor with a custom memory size, custom I/O logic, and unique hardware accelerators without using any additional PCB space. Custom silicon also allows designers to create unique designs that are more complex to reverse engineer.
In the case with Microsoft, the Azure servers' processors may contain functions that are going unused or may require hardware accelerators that are not available. Large amounts of server processing may be involved with data lookup and connection handling. Thus there is a bigger need for simpler cores that can operate simultaneously.
Therefore, Microsoft could create a custom server processor with a reduced number of instructions but has hundreds of network cores that consume less power. Furthermore, the design could integrate specialised cryptography accelerators that can encrypt and decrypt on-the-fly to improve server response times.
Why have companies like Intel fallen behind?
Companies like Intel and AMD remain critical for processors and other related hardware, but that importance is beginning change thanks to the development of low-cost computing and cloud processing.
Recent events at Intel (their inability to break the 10nm barrier), have caused them to fall behind AMD as Intel traditionally only updates its architecture when it updates its node technology. However, while this has resulted in AMD overtaking Intel, it is not the only driving force seeing companies such as Apple and Microsoft turn to custom silicon.
The need for more powerful processors is reducing as more users switch to cloud computing and basic everyday tasks. Powerful processors are important for gaming, workstations, and heavy data processing. Still, even in these applications, RISC could take over, and the introduction of hardware accelerators could remove the need for CISC. The question that now remains is how will AMD and Intel change their future business plan if custom silicon becomes the dominant technology?
Read More
