TI Keeps AI on the Edge Targeting Accessibility at Half the Power Consumption

Artificial intelligence (AI) is poised to offer significant benefits to the consumer by bringing improved performance, usability, and functionality to everyday devices. One space specifically where AI has the potential for important change is in human-machine interaction (HMI).

 

A high-level overview of an HMI. Image used courtesy of Vicky Valla

 

However, bringing AI to the consumer lies heavily in enabling AI on the edge so that everyday devices, no matter how small or low power, can benefit from these developments. With this in mind, companies are pushing heavily to make AI more accessible at the edge with new hardware that supports lower power computation.

Recently, Texas Instruments released a new family of edge AI processors aimed at HMI applications. This article will look at how AI can augment HMI and how TI’s latest products aim to enable AI-based HMI at the edge.

 

Shifting Towards Better HMI Designs

Over the past decades, the way humans interact with devices has changed significantly from bulky computers with command-line interfaces to technologies like the graphical user interface and the touch screen. Today, much of HMI centers around user experience (UX) and user interface (UI) design, with the goal being to make aesthetically pleasing and intuitive interfaces for users to operate. 

 

An example of how AI HMI systems work with human interactions. Image used courtesy of Presans

 

With the proliferation of AI and machine learning (ML), the next generation of HMI is poised to bring entirely new ways of interacting with machines and devices. Already this is being seen with technologies like digital voice assistants (e.g., Alexa, Siri), facial recognition in smartphones, and gesture recognition in AR/VR glasses.

As AI/ML technology continues to progress, it is anticipated that new, unprecedented ways of interacting with devices are on the horizon.

 

Low Power for the AI Edge

Making AI (both for HMI and in general) more accessible to the everyday consumer is becoming increasingly important to enable AI on edge. This way, the AI/ML that powers future HMI applications can live in the device itself, allowing for lower latency, faster response times, and more natural/intuitive experiences for the user.

Here, the most significant challenge from a hardware perspective is how to design devices that can support the data and compute-intensive workloads that AI and ML offer while keeping power consumption at a minimum. 

Unfortunately, these two desires tend to be contradictory. The conflict has forced designers to explore new hardware architectures and techniques to achieve high performance, lower power, cheap, and space-efficient compute.

 

TI’s Newest SoCs—the Sitara AM62X Family

Hoping to make the AI edge for HMI more accessible, Texas Instruments made headlines when they announced a new family of system-on-a-chip (SoCs). The new product line, dubbed the Sitara AM62 family, is a family of highly integrated SoCs that feature several key components that make it particularly well suited for edge HMI. 

 

Block diagram of the AM62 family. Image used courtesy of Texas Instruments

 

From a hardware perspective, the AM62 family is built around a 64-bit, 1.4 GHz Quad-core Arm Cortex A53 processor subsystem, with each core supported by 32 KB of L1 DCache, and 512 KB of shared L2 cache. 

This high-performance processor is matched with a 400 MHz, single-core Arm Cortex-M4F MCU for general purpose usage, a dedicated 3D graphics engine, and an R5F core for device resource and low power management applications. 

Uniquely, the AM62 family also features a dedicated display subsystem that features dual display support, allowing users to bring their edge AI and their HMI control on the same piece of hardware.

According to the AM625 datasheet, the family brings all this functionality without sacrificing power, achieving suspend states as low as 7 mW and deep sleep modes at <5 mW. According to TI, the processor family can bring power savings as large as 50% to industrial applications and can support applications running on AA batteries for over 1000 hours.

Overall, this new family of SoCs could open the door a bit further for more AI edge HMI applications. It will be interesting to see what technology trickles in using this technology. 

 

Featured image used courtesy of Texas Instruments

 

---

 

Interested in other AI news? Read on in the articles down below.

Intel’s AI Processors Aim to Upheave NVIDIA’s Processing Domination

Meta’s Supercomputer Zeros in on Training AI Models for Areas Like Computer Vision

AI Meets Quantum Technology in New Google Spinoff, Sandbox AQ