On the Apple M7 Motion Processor
When Apple announced the iPhone 5s last week the biggest mystery was the M7 motion processor—it may have been the only unknown, as other key internals like the fingerprint sensor and possibility of a 64-bit A7 had leaked. This is interesting, showcasing that the component layer is Apple’s last stronghold of total secrecy.
The M7 is a dedicated processor with one task—to collect and process sensor data. Having an always on co-processor that consumes a tiny amount of power is a trend in mobile (the Moto X does this with other tasks too). It’s better to put the big chip (A7) in deep sleep and keep the little one going.
The M7 likely uses an extremely low power ARM core. A sensor hub connects the processor to all the sensors (i.e. gyro, accelerometers, pressure sensor, magnetic sensor). The communication standard is something simple like I2C between the sensor hub and the processor. It’s unclear who makes the sensor chip until tear down but it’s likely Atmel or STMicroelectronics.
The biggest question around the M7 is what cost / benefit will it provide to overall efficiency (performance per watt) for applications needing location / sensor data?
The M7 will actually enable massive power savings—orders of magnitude for background data collection, consuming around 1% of what the A7 would according to chip architects I’ve talked to. This will radically reduce power consumption for Fitbit-type fitness tracker applications that need to know you moved 5 miles, but don’t need to know the exact GPS coordinates you moved from / to.
However, one misconception with the M7 is that it will also reduce power in apps requiring positioning / mapping overlays and background location tracking. This is not the case.
This is because granular map-level precision will still require GPS. The M7 knows you are moving but not exactly where. What this means is that the iPhone 5s will still need to boot up the integrated GPS components (these are actually inside the Qualcomm baseband chipset). This drains orders of magnitude more power.1
The emerging question around the high end of the smartphone market is whether Apple can continue to innovate at the component level and get consumers to pay for it (A7 + M7 etc). Apple wants consumers to believe that by building its own chips, its products will have a noticeable edge in performance and battery life against the competition. This is subtle but very important. Advertise it and consumers will pay extra. This is keenly why Apple actively markets its chips, but ignores most other specs. The other open question is which Android-based phones (like the Moto X) will continue to do the same?
It is clearly evident we are at a price segmentation phase of smartphones, and this is more reason why the iPhone 5c looks like a smart product—it doesn’t use any of Apple’s latest components and is much cheaper to produce, extending Apple squarely into the mid-range. This was obviously their strategy all along—they may creep into the low end, but why dive in all at once? There are 300 million consumers in China who will decide between a mid-range and a high end phone.
Interestingly there are also new merchant chips which will enable high end Android phones to accomplish what Apple is doing with the M7—e.g. this chip. But it’s clear Apple is way ahead in having the integrated subsystem + the Core Motion frameworks all working together and ready for developers at launch.
More importantly, it’s things like the M7 that showcase Apple’s prescience in knowing how to lead in system efficiency, gains which in sum will continue to help iOS outpace Android in performance per watt by 6-18 months.
And from a more general angle, the industry lesson here continues to be that integrated software / hardware is allowing the top vendors to differentiate—smartphones are simply not following the same path as PCs and it’s really no surprise every big player is vertically integrating.
The reality is that mobile is no longer a question of whether smartphones will condense to a winner-take-all market or follow the path of computing platforms past—they won’t. The real question as the market for low, mid and high end stratifies is what percentage of consumers and which geographies will pay for differentiation.
Today’s power drain for GPS is significant—around 25 mA (at a 1.8V supply) at full power when a satellite fix is obtained (more when searching for satellites). And because new satellite constellations are always being added—e.g. Beidou (Chinese) and Galilieo (EU), power in GPS subsystems will not be going down much over time, even with process shrinks. There are low power modes such as PMM/trickle power that turn the GPS radio off periodically to lower the power. But these only help a little—the fact is GPS draw a ton of current to fix on satellites. ↩