We look forward to presenting Transform 2022 in person again on July 19th and virtually from July 20th to 28th. Join us for insightful conversations and exciting networking opportunities. Register today!
Qualcomm is no stranger to the world of artificial intelligence (AI) on its silicon. But for the most part, Qualcomm’s AI efforts have been somewhat bespoke and specific to certain parts of its tech stack.
That situation changes today with the announcement of the Qualcomm AI Stack. The new offering aims to be a unified platform that brings together Qualcomm’s various AI software tools into a single stack that spans the entire portfolio. The goal of the Qualcomm AI Stack is to continue to meet the specific needs of the various industries that Qualcomm serves and to make it easier for developers to benefit from a single stack.
In a briefing with press and analysts, Ziad Asghar, vice president of product management at Qualcomm Technologies, explained that some companies are looking to deploy AI across multiple product lines. With the Qualcomm AI Stack, Asghar says, a company can take an AI model designed for one industry — say, a mobile use case — and then use that same model for another deployment, such as an edge network or even a deployment in the automotive sector. In the past, this might have been a challenge as the different use cases rely on different Qualcomm hardware, which previously didn’t all support the same AI stack.
“We believe that with Qualcomm’s AI stack we will support developers and OEMs [Original Equipment Manufacturers] to be able to do so much more with the AI capability we’re building into our devices,” Asghar said.
An AI stack for many Qualcomm productss
Recently, Qualcomm has released a steady stream of hardware and software designed to accelerate AI.
Qualcomm’s mobile chip is the Snapdragon, which integrates hardware acceleration capabilities for AI. The Qualcomm XR platform, on the other hand, is intended for virtual and augmented reality applications that also benefit from AI. The company is also targeting Edge AI and robotics.
“We’ve now made this leap where essentially the same offering can cover every single line of business that we have today,” Asghar said.
The overall market for AI-optimized silicon is large and growing. According to Verified Market Research, the AI chips market is expected to reach $202 billion in revenue by 2030, which represents significant growth from the $10 billion in revenue in 2021.
Qualcomm AI engine direct
The Qualcomm AI Stack provides support for multiple AI frameworks, including the open source TensorFlow PyTorch and ONNX. The tier below mainstream AI frameworks is where Qualcomm AI Engine Direct technology now comes into play across the Qualcomm portfolio.
Asghar explained that the Qualcomm AI Engine Direct is for developers who want to benefit more directly from the silicon capabilities that Qualcomm hardware enables. AI Engine Direct directly runs software compilers, programming languages and math libraries, as well as profilers and debuggers. AI Engine Direct also supports multiple operating systems including Android, Windows, Linux and even QNX for the automotive industry.
The ability to support the various functions developers may need to enable an AI application is important, but more is required to successfully deploy and run AI. One of the tools in the Qualcomm AI Stack is the AI Model Efficiency Toolkit, which consists of several components. One is the ability to compress a model from something that might have been trained in the cloud to something that can run on an embedded device in a kitchen.
Ashghar also highlighted Qualcomm’s neural architecture search capabilities developed with Google. He explained that Neural Architecture Search enables higher accuracy and lower latency to run models, even on lower-powered devices.
“I think hardware is obviously critical, but increasingly AI software is absolutely critical,” Ashghar said. “We have made very large investments to be able to produce this AI stack, the best-in-class AI stack for the intelligent edge.”