• Dedicated processors will dramatically expand the use of AI in smartphones with voice assistants being one of the first beneficiaries of ‘on-device’ AI processing
Smartphones have been leveraging the capabilities of artificial intelligence (AI) for some time. However, till now the processing has been done either in the cloud or distributed across the various computer chips in the device such as CPUs, GPUs, and DSPs.
As AI becomes part of the mobile experience, smartphone system-on-chip (SoC) vendors are racing to improve the machine learning capabilities of their chips by integrating dedicated AI processing cores into their designs. The key benefits of this are higher AI processing performance and lower power consumption. However, this must be balanced against the actual need for AI processing, which until recently has been limited.
At present, AI is being used mostly for camera-related applications (facial recognition, image enhancement, and more). However, Counterpoint Research expects to see AI being used extensively for a raft of other applications over the next two years.
“We see voice assistants as one of the first applications to benefit from device-based processing,” said Gareth Owen, Associate Research Director at Counterpoint Research. “Today, most voice processing in smartphones is cloud-based. However, voice assistants will be able to process commands quicker and respond faster with on-device processing. It also resolves privacy concerns,” he added.
Apple and Huawei were the first OEMs to include dedicated AI processors in their SoCs (A11 and Kirin 970 chips respectively) launched in their flagship handsets in September 2017. Two years on, virtually all other SoC vendors are following suit. For example, Qualcomm is offering an AI Tensor Accelerator in the Snapdragon 855’s Hexagon DSP for the first time.
Counterpoint Research forecasts sales of AI smartphones will increase to 1,250 million units in 2022 from 190 million in 2018, representing more than three-quarters of all smartphones shipping in that year.
A key driver behind this forecast is the expected launch of several Android smartphone models using Qualcomm’s 855 SoC in 2019 and beyond. Counterpoint expects that dedicated AI cores will also be included in some mid-range smartphones during 2019 using Taiwanese vendor MediaTek’s new P90 SoC.
Although AI may not be immediately visible to smartphone users today, as the machine learning market evolves and more applications appear, smartphones with dedicated AI hardware will clearly be at an advantage.
“With Qualcomm joining the fray and incorporating hardware-based AI acceleration in its latest Snapdragon SoC for the first time, industry-wide adoption of AI silicon in smartphones is inevitable,” said Peter Richardson, Research Director at Counterpoint Research. “And with Qualcomm keen to extend machine learning capabilities into its lower cost chips, plus the availability of AI SoCs for mid-range phones from other vendors, this trend will quickly extend to mid-range and even low-range smartphones,” he added.
