Processor and GPU cores usually get the limelight, driven by the ARM and Imagination machines occupying the center square of most SoC designs. CEVA has quietly been assembling DSP IP in most of the squares around the edge, and may have just reached critical mass for wearables and IoT devices.
There was plenty of coverage on the recent RivieraWaves acquisition, turning a foothold in Bluetooth into a solid block of squares with world-class Wi-Fi and Bluetooth LE IP. Looking at the CEVA offering now, what they have done with DSP-based IP around the CPU, GPU, camera, and display is impressive.
If that diagram looks a lot like a smartphone, it is – and it isn’t. No doubt CEVA stands as the benchmark for LTE baseband IP, and that has fueled much of their success with DSP IP to date. What happens next, as the smartphone market transitions from hectic worldwide growth to a more controlled pace, is a pivot that CEVA has been laying the groundwork for.
By no means is CEVA stepping away from the smartphone, but we can see an expanded strategy at work here. How do wearable and IoT “things” differ from smartphones? The trends in play:
- There will be no octacore-and-beyond madness in most things, for the foreseeable future – one small, efficient CPU core will likely be it, more like today’s microcontroller.
- Many things will forego the GPU and OpenGL ES entirely and opt for a simpler display driver subsystem, or no local display beyond a couple of status indicators.
- Even more things will skip the 4G interface, either going for a 3G connection or sliding into the smartphone and home infrastructure with Bluetooth LE or Wi-Fi.
- Wi-Fi will become more integrated in SoCs for things and be “sleep hardened”, with power-aware 802.11ac implementations and streamlined IPv6 processing.
- Sensor fusion with algorithms like Kalman filtering becomes much more important, bringing increased use of low-power DSP capability.
- Always-on, driven by dedicated DSP cores performing functions such as voice and facial recognition, becomes mainstream for things.
Considering what is inside the computational photography and computer vision blocks is very similar to the needs for sensor fusion, the CEVA strategy becomes clearer. Applications such as indoor navigation are going to drive new and more complex algorithms that will need to run on smaller devices.
Commentary in the CEVA presentation announcing the Riviera Waves acquisition indicated 25% licensing revenue growth in 2015. However, I think that is the “as-is” case combining the businesses straight up. What I think is really happening here is a jump-start for wearable and IoT design wins that will materialize in two or three years.
It is interesting CEVA is staying on the signal processing side, and not venturing into the MIPI arena for cameras and displays. That may just be sticking to the DSP knitting, and not entering where there are a lot of other participants. However, going out a limb … I think something is going to happen in the industry with the wearable display controller soon.
CEVA is well set up here for anything that filters, correlates, convolutes, transcodes, or similarly processes signals in noise without using a lot of power. That isn’t going to be restricted to the heavy lifting required for LTE or real-time computational photography in the smartphone. New doors will open in wearables, particularly in areas like sensor fusion and aural user interfaces.