WHAT’S AI DOING IN YOUR PHONE?
Huawei’s Kirin 970 chipset claims dedicated artificial intelligence processing, but what does that mean?
THE AI BRAIN
Huawei’s Kirin 970 chipset doesn’t only come with a traditional CPU and GPU. There’s a third processor, dubbed as the Neural Processing Unit (NPU) that’s specifically built for AI-related computing. In comparison to the CPU found on the same chipset, this NPU:
• Offers 25 times the performance at half the CPU’s size
• Consumes just 1/50th of power
Huawei uses all three processors (CPU, GPU, and now an NPU) in their latest Kirin 970 chipset. It does so by handing off graphical rendering to the GPU, memory allocation and tasks scheduling to the CPU, and finally, AI computing to the NPU alone. One example of an AI feature is image recognition (pictured). Should Huawei correctly implement this part of the NPU, it can result in more accurate smartphone photography, i.e., automatic color-correcting for a human male with his brown pet dog.
It’s worth noting that the NPU does not have a machine-learning nature – Huawei specifically said that the AI is preloaded into the processor, and it uses its programming to do what it must. Should the algorithm require fine-tuning or new features, Huawei will provide them as OTA (over-the-air) firmware updates.
REAL-WORLD USES FOR KIRIN 970’S NPU
While it’s not in charge of delegating workload, the NPU has its fair share of tasks cut out, and it spans across various processors on the Kirin 970 platform. Building on the photography example, Huawei’s NPU also assists the image signaling processor (ISP) by:
• Enhancing its response time throughput by 25%
• Increasing its processing response volume by 30%
• Providing dual-channel parallel processing
• Detecting and employing various motions, i.e., static, slow, medium, and fast motion
• Using hardware-based face detection
• Having an intelligent camera scene detection method
These improvements aim to help the photographer take less blurry photos, with reduced chances of missed shots due to the faster response by the chipset (and thus, the phones). The NPU also ‘unlocks’ certain capabilities previously unavailable to smartphones; the user can employ a 3x3 focus-detection area (as opposed to your regular detection square), and the AI can provide the context by reading other elements (background, mood) to give an ideal balance to the result.