exxos's Atari, Amiga & retro forum

All,

I've done some analysis of the Mathesar designed SMPSU modules that have been flying around after watching one catch fire on me during the Amiga Ireland event and another one caught fire on @chucky live last weekend.

Here is the circuit
and here is the reference design for this part.
What we observed was that if the output was shorted to ground the diodes exceeded 150 degrees. In my case started to catch fire. I was uncomfortable with this because linear regulators have thermal shutdowns to prevent fires. Closer reading

7.3.3 Soft Startup
After enabling the device, internal soft startup circuitry ramps up the output voltage which reaches nominal output
voltage during a startup time. This avoids excessive inrush current and creates a smooth output voltage rise
slope. It also prevents excessive voltage drops of primary cells and rechargeable batteries with high internal
impedance.
The TLV62569 is able to start into a pre-biased output capacitor. The converter starts with the applied bias
voltage and ramps the output voltage to its nominal value.

Basically these diodes added by Mathesar are screwing up the devices ability to shut down and have bypassed all safety. They are really really dangerous. Do not leave machines with these devices in on unattended. They could cause a fire. Just dont do this.

I believe they are there to prevent 060s from getting 5v on input pins before 3.3v is up. This is the wrong way to do power sequencing. Use an SMC or put buffers between all your pins and the 5v inputs. At the very least put a resistor in to limit the current through those diodes. Jeez.

The TF1260 does not need a this as no 5v rails go direct to the CPU. Every pin is buffered through a Bus Switch or the Xilinx CPLD. Only those devices ever see 5V. For the XIlinx " External I/O voltage may not exceed VCCINT by 4.0V.". The soft startup on those PSUs is 800uS. Let's assume it's linear (it's totally but ignore that for now). That means it takes 242us to get to 1V. The A1200 has a 1000uF cap on the 5V input, from the schematic. For it to charge that cap too fast, it needs to get at least 4 amps into it, not counting whatever the rest of the board is doing. So there is absolutely no need for those diodes.

In anycase I'd rather replace a couple of CPLDs than someones house.. JUST NO! Dont do this!

I've seen that done before, its because some chips cannot start up under load. So the diodes may kick it to 2V kinda thing to get things started.. but your right, if the output is shorted,

EDIT:

My PSU's I use ones where I can set the soft-start current manually.. So mine start up with 1amp tops, they dont just ramp up to 8amps right away.. Under short conditions the reg wont increase voltage so it protects the circuit. If they shorted out, it enters hiccup mode and limits the current..

Though diodes like that in series just isn't the way sort-start is supposed to work.. If the chip can't start up under load, then get a better chip, or one with manual soft start and set it yourself.

exxos wrote: ↑Wed Feb 10, 2021 8:44 pm I've seen that done before, its because some chips cannot start up under load. So the diodes may kick it to 2V kinda thing to get things started.. but your right, if the output is shorted,

But the diodes dont even conduct until there is 3 x the diode drop across them... This chip seems to be ok with load. So we're really talking about the lag between the SMPSU output and the 5V input. Which is 800uS and the gap will be unlikely to ever be more than 3v at a time.

The main issue is that people read what is written in the Motorola docs and take that as gospel. Forgetting that documentation is always massively conservative and occasionally fictional. I've worked as an engineer for 20 years now and a good deal of the the time the docs are elaborate works of fiction or document how people would like the system to work (rather than how it actually does). Sometimes off hand remarks from engineers end up in big warning boxes in docs. Things get lost in translation and sometimes engineers are just wrong.

I'm willing to bet a 68060 CPU will survive being put on a full 5v bus with no main power supply at all there. I'm also prepared to sacrifice one of those chips freely rather than kill people.

terriblefire wrote: ↑Wed Feb 10, 2021 9:24 pm But the diodes dont even conduct until there is 3 x the diode drop across them... This chip seems to be ok with load. So we're really talking about the lag between the SMPSU output and the 5V input. Which is 800uS and the gap will be unlikely to ever be more than 3v at a time.

If the diodes were 1V drop each, on 5V, that's 2V that the 3.3V rail will get when 5V reaches 5V.. If anything it would bypass the chips softstart as the 5V rail ramping will equate to 2V on the 3.3V rail on power up.. as to if the chip even likes being run like that.....

When building and testing 3660s. I usually put in a header for the regulator. so I can use 040 or 060. so I can put on one DCDC to get 3.3V to test 060s..

MANY times I have fotgot to put in the 3.3 board.. even for minutes measuring if I get data until I notice "DAMN no 3.3V". been on for even 1-2 MINUTES. no dead 060.

That chip looks to struggle after 2amps, what does the 060 even pull when overclocked ?

This topic begs for a stream:) just saying...

exxos wrote: ↑Wed Feb 10, 2021 9:37 pm That chip looks to struggle after 2amps, what does the 060 even pull when overclocked ?

about an amp @50mhz with the modern SDRAM on the chain. 1.3A at 75mhz ... i havent measured higher.

adam wrote: ↑Wed Feb 10, 2021 9:40 pm This topic begs for a stream:) just saying...

I am considering coming out of YT retirement to set something on fire.

EDIT: But lab is not what it used to be due to house sale

No worries, be back when you are ready and armed.