24-02-2022 | By Robin Mitchell
The release of the Apple M1 took the world by storm, and since then, it has demonstrated that custom silicon-based on ARM cores can compete with mainstream CPU technologies. What made the Apple M1 so unique, what advantages do SoM and SoCs present, and why do the Apple M2 rumours demonstrate the success of SoM designs?
The Apple M1 is essentially a System-on-Chip (SoC) mounted onto a System-on-Module (SoM) that utilises an entirely custom architecture devoted to running Apple products, including macOS. At the heart of the M1 are 8 ARM cores, with four being dedicated to high-performance while the other four are dedicated to high efficiency. Also integrated into the SoC is a GPU with eight cores and multiple levels of cache.
This SoC is mounted onto a board that integrates the system memory with either 8GB or 16GB options, and these are not upgradable. The entire device is then encapsulated in a heat spreader, and connectors on the underside of the board allow for mounting in devices. For this reason, the Apple M1 is more of an SoM than an SoC as it combines multiple silicon devices into a single module that is then used in an external circuit.
SoCs and SoMs are emerging technology, but they are not “new”. The Raspberry Pi is an excellent example of a product that introduced the world to the many advantages of SoCs. The Broadcom chip that powers the Raspberry Pi integrated the CPU, MMU, GPU, and I/O controller in a single package. This meant that an entire computing system could be integrated into a single credit card-sized PCB. SoCs have also been heavily used in IoT designs, with the ESP32 being a good example; it integrates the CPU, memory, and radio controller all in a single package.
So why was the M1 so ground-breaking? Until the M1, only mobile personal computing devices such as tablets and smartphones used ARM cores. Mainstream computing devices (PCs and laptops) used x86/x64 architectures, and only laptop devices used mobile versions of mainstream CPUs. These CPUs would be sourced from essentially one of two companies: Intel or AMD.
It is well known that Intel and AMD are somewhat lacking in the mobile industry, and having a powerful mobile processor often means high energy usage meaning reduced battery life. The energy savings afforded by ARM cores is why they have remained a key player in mobile devices.
But the Apple M1 changed all of this as it was designed with both mobile and desktop processing in mind. The use of multiple cores, each dedicated to either high-performance or high efficiency, allows for a system that can optimise performance while saving power. This efficiency is far better than simply clocking down a CPU or throttling memory as high-performance cores dedicate their silicon space to high-performance while high-efficiency cores dedicate their silicon space to high-efficiency.
This means that during times where efficiency is needed, the only hardware that is dedicated to this task is in operation. Empirical results show that the Apple M1 consumes 39 watts when operating at maximum load and 7 watts, considering that the equivalent system using an Intel processor consumes 20 watts when idle and 122 watts when under load. The result of the M1 has effectively doubled the battery life of Apple products.
By far, one of the biggest advantages of SoCs and SoMs is that all silicon space used is dedicated to the specific application that the device will be used in.
For example, a modern CPU from Intel or AMD has to be a generic computing machine as the application for that CPU is not known. A customer could just as easily use an Intel CPU in a robotic controller where I/O must be given priority, for use in a supercomputer where floating-point operations are essential, or a generic laptop where energy efficiency must be considered.
This means that Intel must make their devices as generic as possible to be easily integrated into any application. However, this makes such a device more of a “Jack of all trades, Master of none” as it cannot be specialised for any one task. A SoC, however, can be designed to only integrate hardware essential for the task that it will be used in. For example, a mobile processor could have low-energy circuits used for energy consumption, while a vision system in a self-driving car may integrate AI modules for faster processing of neural nets.
Thus, SoCs and SoMs allow designers to customise a design that fits the application perfectly.
While reports have not been confirmed, it is widely accepted that Apple is working on the M2, which will replace the M1. Some rumours suggest that the cores will not see dramatic changes, but the GPU will be improved, and CPU frequencies will be increased. The supposed release date for such a device would be late 2022/2023 and power most Apple products.
Even though the M2 hasn’t been confirmed by Apple, the success of the M1 perfectly demonstrates the advantages of SoCs and SoMs. Of course, the Apple M1 isn’t the most powerful processor on the market. That crown belongs to the Ryzen ThreadRipper 3990X, but when it comes to performance per watt, the M1 is king. The high efficiency of the M1 allows it to massively extend battery life, while the ability to tailor the hardware just for macOS allows for hyper-efficient OS design.
There is no doubt that SoCs and SoMs will become the industry standard for computing devices. But what would be exciting for engineers is a custom chip service that could allow customers to choose semiconductor dies and have them connected together on a SoM, just like a custom PCB service with component assembly.