MCU trends

ARM announced Cortex-M55 this week, also with vector processing for machine learning & signal processing. It's almost certain to become Teensy 5.x... in 2-3 years when chips become available.
 
Priorities in that domain are obviously changing. Almost nobody talks any more about improving the analog building blocks in mixed signal MCUs - it's rather the opposite: Powerful integrated ADCs and DACs tend to be eliminated or externalized. The Teensy 3.2 had 2 16bit ADCs with 12 or 13 useable bits resolution and one 12bit DAC with useable sampling rates slightly above 500kHz. The Teensy 3.5 and 3.6 had even two of these DACs. I had hoped for >1MHz bandwidth and at least 14 if not 16 useable bits resolution on 4 ADCs and 4 DACs in their successors, but we got a T4 without any DACs at all and with degraded ADC resolution.
That's highly frustrating for old farts like me who still dream of full universal analog (from DC to video signal range) and digital integration on one single chip. Time for me to retire?
 
You don't want the noise from all the other circuits on the chips. External circuits can do this much better and are not that hard to use.
Maybe some day a company makes a easy to use addonboard for old farts like us. For Audio, there are plenty.
Then, more and more becomes digital. The use-cases are not that much. Video is digital nowadays, too. No need anymore for ntsc, fbas, vga and so on. And audio - everything between in and output is digital anyway.
 
Last edited:
I too had hoped for DACs. Believe me, I've repeatedly begged for DACs from my contacts within NXP. We are getting 1 DAC in the next chip, which is a lot better than none, but I'm pretty sure it's not because of anything I said. I'm pretty sure they decide the features based on requests from only very large customers and maybe by looking at what ST is up to with their STM32 chips?

There are 2 unfortunate trends affecting analog. It's well known that as mosfet channel lengths and other silicon process features shrink, the speed and amount of digital circuitry scales up, but the quality of analog circuitry scales downward. The same is true for flash memory, which is why use of external QSPI flash chips is becoming the norm with these fast parts. But memresistor technology may become possible in the future as things keep scaling even smaller...

The other trend looks like human factors. My best guess is iMXRT was mostly designed by NXP's teams who've worked on the iMX application processors, rather than the people with experience from Kinetis, HC08 and other microcontrollers. Some of their analog design decisions, like not bringing VREFH to a pin, seem downright negligent. Specs for the next 1170 chip are still under NDA, so I can't comment on details, but I will say it does look like they're learning some lessons in analog.

But the dream of a chip with very a high speed CPU and high precision analog on the same silicon die runs counter to the reality of semiconductor physics. As silicon scales to smaller feature sizes, only digital circuitry tends to reap the benefits.
 
I agree that it is nearly impossible to get good results from onboard ADCs in the newer processors. For me, good results means 1-2 bits LSB noise at 16-bit resolution. I've gotten near those results on the Teensy with two different external ADCs: The LT1867L from Linear and the AD7606 from Analog Devices. In each case, I kludged in a board designed for another logger. The AD7606 is interesting because it simultaneously samples 8 channels and has some built-in low-pass filters. I was using it for experiments in audio beam forming with an array of simple microphones.

Another argument for external ADCs is that most real-world logging systems will need low-pass filtering to avoid aliasing. If you can sample fast enough, you can use digital filters to prevent the aliasing, but that increases power dissipation. In one logging system with the LT1867, I used Sallen-Key low pass filters which attenuated about 24db at 200Hz, sampled at 400Hz and applied a 30-pole FIR filter that attenuated about 60dB at 60Hz. The results were good enough to measure seawater temperatures with milli-degree precision at 50Hz. If you want really good analog results, an external ADC and filtering are necessary and ought to be on the same board with it's own voltage regulation. This is especially true if you are going to write data to an SD card as it is acquired.

Between 1998 and 2012, I designed and sold a range of analog peripherals for an earlier SBC, the Persistor CF-2. I'm considering adapting some of those boards to work with the Teensy 3.X and 4.0. The design, PCB layout and testing would be interesting, but I'm way too retired to be interested in marketing the boards.
 
Waiting for Teensy 5 with Helium support

Consider ARM announced Cortex-M7 in 2014. A couple years later, Atmel, Freescale & ST released expensive, power hungry M7 chips running between 200 to 300 MHz. It wasn't until NXP made iMXRT in 2018 that we got a really fast M7 chip. Teensy 4.0 development began in early 2018, beta testing began in January 2019 and it was finally released in August 2019.

Cortex-M55 was announced only 12 days ago. There's a chance things will move along faster than M7 did, but odds are we'll be waiting at least a few years for a really capable M55 chip.

Hopefully your project can wait til 2024-2025 time frame....
 
Oh yeah, the project is for the long run and I expect there to be few iterations of T4 before T5 hits the market.
 
Back
Top