Imagine an affordable MCU capable of performing high-end AI/ML tasks
By Clive “Max” Maxfield, Editor at DENA.
I remember when the early 8-bit microprocessor units (MPUs) appeared on the scene circa the 1970s. Names like the 8080, Z80, and 6502 will stir fond memories in many of my peers. Similarly for early 8-bit microcontroller units (MCUs) like the 8051.
The idea behind MCUs is awesome—that is, to have a single-chip processor with on-chip memory (both non-volatile and SRAM) and on-chip peripheral functions. A device that can perform its role in life with only a handful of low-level support components.
Until recent times, MCUs have been largely relegated to performing relatively low-level tasks, like monitoring sensors and controlling actuators.
These days, of course, everyone wants to incorporate artificial intelligence (AI) and machine learning (ML) capabilities into almost every product. In response, many companies have introduced MCUs that can support simple time-series AI/ML use cases like monitoring vibration data, detecting anomalies, and making predictions along the lines of, “This machine will fail within the next 72 hours.” Until now, however, small, affordable, low-power MCUs capable of handling more advanced use cases like voice processing and even video have been thin on the ground (some might say “non-existent”).
I must admit to having a personal interest in this. One of my current hobby projects is to create a Countdown Timer that can display the time remaining to the commencement of my 100th birthday celebrations at 11:45 a.m. British Summer Time (BST) on 29 May 2057 (mark your calendar). I’m using 14 steampunk-looking vacuum fluorescent displays (VFDs) to display the time in Years (YYYY), Months (MM), Days (DD), Hours (HH), Minutes (MM), and Seconds (SS).
However, if we say there are around 33 years until we all don our party hats, then this equates to around 290,000 hours, but VFDs have a typical life expectancy of only around 10 years or ~90,000 hours. I’d feel foolish not knowing when the time came to blow my party horn.
What I want to do is mount a webcam on my Countdown Timer, feeding its signal into an AI-MCU that can detect when anyone is looking in its general direction. If no one is looking, then we can keep on counting but power-down the VFDs. It’s only when someone starts looking that we turn the VFDs on.
Well, this all sounds simple enough. All we need is an appropriate AI-MCU. Happily, a new player called Alif Semiconductor (www. alifsemi.com) has joined the party (no pun intended). Alif’s Ensemble MCU boasts dual application processor cores, dual real-time processor cores, an AI/ML neural network core, a graphics imaging core, up to 13.5MB of on-die SRAM, and up to 5.75MB of on-die non-volatile memory (NVM) in the form of magnetoresistive random- access memory (MRAM).
I’ve reached out to the folks at Alif, and I’ll keep you informed as to the progress of my Countdown Timer project in future issues of DENA.