Adventures in TinyAVR 1-Series

Yeah you’re probably right, I’m overthinking this :person_facepalming: . I just saw a capacitor in the feedback-loop and thought hmm, I’ve seen that before in opamp-configurations, but I did not think about the other things on the input/output…


Perhaps it’s possible C4 is there to not “overload” the opamp-output if the gate (capacitance) of the mosfet gets charged? Although R6 already limits this current charging/discharging current.

R7’s function (because of the rather high and pretty common value) might be to make sure the gate get’s pulled low if the OPAMP is disabled (high-impedance output instead of a low/high?), and so to make sure the mosfet is open.

R8 (because of its common value) might be there to limit the current flowing in -V (which is very low but not totally non-existend) and so not mess with the voltage over Rsense (~high-impedance voltage measurement, a multimeter for example has an impedance in the MegaOhm-range).

Capacitor C4 is needed for AC frequency compensation of circuit because we don’t have any negative feedback and we will have open loop amplifier circuit which will oscillate very easy.

Yeah that’s what I roughly got after a bit of reading, and unsurprisingly it is also necessary when using the same current control loop with a switching converter.

edit : thanks for the link.

Some corrections: RC filter of that PWM DAC is R3C3 not R4-C3. I don’t like that one stage filtering. It is very slow and noisy in most of time. Also that slow down reaction of CC driver of input signals especcially in strobes. Better choice is to be used at least dual stage RC filter. Also another problem is heat from mosfet. Most of time at high currents and low Vf LED its need to dissipate a lot of wasted power and it is possible to overheat. That is major drawback for linear driver.

Ah yes you’re right for the RC filter.
About it being slow, I’m not sure this is really an issue with 2pKhz PWM that we can do on the TCD counter at 10bit on those 1 series. Personally I don’t care about strobe in the slightest, but I guess some people do.

The advantage about linear drivers is that they’re quite simpler than switching topology, low BOM and board space, and if the feedback loop is done by the MCU instead of the OpAmp by reading the sense voltage by the ADC, like on the Unicorn 1.0, then the part count is ridiculously low. On this driver it consists of the RC filter, the sense resistor and the mosfet. 4 parts + the MCU ! It is then trivial to add another 4 low power/small parts for a lower current channel to achieve very low current.
Having this implemented in FSM/anduril would be great, but I know nothing about this, I don’t know if this is complex or not.

On the same light there is thermal compound between the mosfet and body to help thermal transfert, adding thermal pads might be a good idea, doesn’t help the efficiency though.

Alright, here’s my first driver, actually first PCB I’ve ever designed. I started with a simpler linear driver to familiarise myself with Kicad :

Not tested

Features :

- 22-20mm, 18mm GND ring inner diameter.

- Constant current Linear 5A, can probably do more with better dissipation, can be configured for less of course.

- One sided, all 0603 passives (except one 0805 sense resistor) for ease of fabrication.

- Very low moonlight.

- Attiny1616 VQFN-20 or others compatible, 3 wires flashing.

  • RGB AUX LEDs

Here is the schematics :

I explained the solution I came up with for moonlight on the Lume X1 thread

Basically with a 10mΩ sense resistor for example, the sense voltage becomes too low as we decrease Vref to reduce the LED current, with 10bit it goes down to 49μV (50mV/1023) on the lowest level, at this point even with a very precise OpAmp there is (probably) noise which makes the output flicker. We don’t want to use a bigger sense resistance as it would reduce efficiency and need a high power resistor.

The idea is to switch to a higher sense resistance for low currents, we can dim with 1:250 or 1:500 ratio that doen’t produce flicker, for example from 5A (50mV) to 10mA (100μV) with the high current range (10mΩ sense resistor), and then from 10mA to 20μA with the 5Ω sense resistor. We switch between the two sense resistors with a very low RdsON NMOS (Q2).

Simulation example :

Red is current and green is Vsense.

At first Q2 is ON so we’re in the high current range, then we dim to 10mA At 9ms Q2 is turned OFF to switch to the low current range, the reference need to be put back to 50mV, we also get 10mA Then we can dim to 20μA. Q2 RdsON needs to be very low as to not influence the sense resistance too much (Q2 is part of it). The mosfet chosen has a RdsON arround 2.2mΩ at 2.8V (Logic voltage). Due to RdsON variation (parts, temperature), the current won’t be as precise as it would normally be, but should still be OK for our application. We could use a gate driver with charge pump to drive the mosfet with a high Vgs and reduce its resistance but it increases BOM and board space.

Before I publish it on Oshpark I’d like people check that I wired the MCU properly, I wired the current control (PWM_CC) to a TCD pin since it can do higher frequency at lower power consumption. Also if anybody see any mistakes, or has comments for improvements.

Regarding Oshpark fabrication limitation, it seems like the copper can’t be closer than 0.4mm to the edge, so I added that much to be filed down, kind of annoying though.

The code will need some adaptation to switch between the two ranges, I imagine it’s not very complicated but I’m kind of dreading that part as I’m quite bad with this :weary: I hope that Gchart or somebody else can help with this.

Parts :

- LDO : 2.8V, this one only have reverse batt protection but not reverse current protection thus it doesn’t protect the MCU from reverse voltage, I haven’t found a 2.8V with both, MIC5225 has both but only 2.7V, will increase Q2 RdsON a bit. I don’t really want to use a diode due to the voltage drop.

- Q1 : needs a lot of margin for the safe operating area so it doesn’t blow (mosfets don’t like being in the linear region), this one is a bit expensive though, can use a PSMN2R4-30MLD instead, it seems like a popular reference.

  • Q2 : lowest RdsON I could find at Vgs=2.8V in a 3.3x3.3 package, not exactly cheap.

I was also working on a boost driver based on the TPS61288, with the same multi sense resistor topology for very low moonlight, but Loneoceans showed his driver first, in his thread he said that he is using the same solution, so the good thing is that it confirmed for me that this is a viable solution. In his driver I am pretty sure he is using 3 sense resistors and 2 mosfet to achieve an even higher dynamic range!

Looks very nice! Here’s my thoughts after looking at it for about 1 minute:

  • With Oshpark, I run my copper to the edges without any problem. I wouldn’t worry about leaving a margin and then filing it down.
  • It seems like 0805 is kinda small for the main sense resistor? It seems like we usually have 1206 sense resistors due to the amount of current flowing through them.
  • A few months back I switched the order of the flashing pads, putting Gnd (V-) in the middle. It makes some layouts easier, plus if you reverse the programming key on accident, I feel like you’re less likely to fry your MCU from reverse voltage. Here’s a link to the updated programming key. And a pic of a recent driver using that layout:

Thanks.

  • So I can put the edgecut directly at the final diameter ?
  • I think I saw some drivers with similar current with 0805 sense resistors, it would need to dissipate 200mW in the current configuration, I saw power rating of 500mW and even 1W for 0805. We can also reduce the resistance a bit because now we have so much dynamic range and we can just use 1:250 dimming, not too much though because then the mosfet’s RdsON would influence the sense resistor too much, though some actual testing would be needed here.
  • I will change that, pitch is 0.127 1.27 mm (1/20inch) ?
  • Yup. That’s how I always do it at Oshpark and they come out just fine.
  • That’s fine then, no problems.
  • Sound good. And yeah, 1.27mm spacing (center-to-center). I use via’s (0.4mm drill with 0.75mm outer diameter), but it’s not necessary. They just help keep the pins in place while you’re flashing.

So many components … :slight_smile: , here how looks linear Infireon driver from Geko headlamp
Here Avsel linear

I talked about it 5 posts above, the control loop is done by the MCU so no need for an OpAmp and all its passives, but this is entierly software and I don’t have the skills necessary, nor do I really want to learn them, and as far as I know Inferion code isn’t open source, if it was it would certainly be easier for people here to implement it in anduril.

Aside from that it can’t do low moonlight, they could add 4 small components costing a few cents for a second channel but they don’t I don’t know why. By the way this new headlamp looks really bad, maybe they are just really cheap.

Anyway I mostly made this linear driver as an exercise and also to propose something not too hard to build, moonlight capable, and a step up from 7135s. Next I will do boost and buck drivers because I like efficiency.

Edit : Avsel one is similar to any linear driver with OpAmp, except it doesn’t have a LDO, I don’t know why, can the MCU output a fixed voltage PWM no matter the Voltage on the VCC in ?
It doesn’t have moonlight levels.

There are MCUs on the market with integrated op amps, logic blocks, and some other peripherals. For example, some of the very low cost 8 bit PICs have op amps, likewise the powerful PSOCs and MSP430s etc (some with similar or lower price points than the AVR 8-bit MCUs). As far as I am aware though, there are no Attinys with integrated op amps though the 1616 does have some analog comparators and a small amount of programmable logic. That said, thefreeman's design probably offers some additional benefits such as Anduril and better control of the analog loop with component choice and feedback optimization. Even if constrained (at least for now) to the Attiny series, Anduril itself is worth the extra cost. thefreeman, nice job, always great to see more options with new designs . As a side note... the pandemic is really affecting supply chains around the world. Many of the QFN AVR1-series MCUs are sold out or very short in supply at the moment, with a lead time of almost 3/4 a year..

Holy crap, I new the microcontroller market was getting rocked but I didn’t know it was this bad. I just check attiny1616 stock at the usual places and they’re ALL GONE. Scouring Octopart now… :frowning:

Edit… well, I think I’ve got 10 on hand which will last me a little while. I just went ahead and ordered 25 at Mouser, they’re expecting more sometime in July. If I can’t get them now, I might as well at least be in line for some.

I planned to order at Farnell but that’s in UK, do they ship to US at reasonable cost ? Indeed stock has disappeared at Mouser, last time I looked there was stock :open_mouth: , I should hurry to make my order.

I added the suggestion from Gchart :

- only one via for the flashing pads because I couldn’t fit more without moving a lot of things

- gnd in the middle of the pads

  • edgecut at 22mm

Thanks :smiley:

Yeah I don’t exactly know how it is done in the Unicorn 1.0/Geko, I thought I read that Vsense is read by the ADC, with some wizardry to increase resolution, then a PID loop to control the current, not sure though, maybe it’s something else.

I saw those at Farnell and some over in Asia. Every place wanted like $40+ in shipping fees to get them here - no thanks! Well, I guess Farnell “only” wanted £25 in shipping. I guess I’ll wait for my Mouser order this summer. I’ll try to make my 10 on hand stretch out until then.

That Geko looks like a decent ZL clone - except if there’s no moonlight there’s no point cloning ZL (to me anyway)

I’ve been away for a while… nice to see that your work with 1-series is catching on.

Maybe I’m being a stickler, but the 1 series doesn’t need three flashing pads. It’s needs GND and UDPI, and the MCU needs to be powered. If I don’t have the driver powered already I just connect V+ from the Xplained Nano to the battery spring with a crocodile clip.

Another thing… you mentioned shortage of 1616s and you see them on Farnell UK. The US branch newark.com has them in stock (selling fast though), do you still have to pay that much postage if you order through Newark? They do refer to UK stock but you might be able to get around the shipping? I can’t order from Farnell UK directly myself, but in Sweden we have a components shop that allows anyone to order from Farnell UK and only pay local Swedish postage from the Swedish shop. I just got a bunch of 1616s delivered. Perhaps there is a similar service in the US for Newark if you can’t order from them directly?

I think when I tried Newark, it was quoting me outrageous international shipment fees since the stock was going to come from UK.

I agree, you don’t need VCC if the MCU is powered, but what if you want to flash a driver that isn’t installed in a flashlight? Or it’s not able to be powered when you have access to the flashing pads (most flashlights)? In my mind, it’s no more difficult to include all 3 pads and it gives you more flexibility. Just my 2 cents.

Hmm, and here I was wondering if it mattered if the V+ for flashing was connected before or after the LDO, if like, the MCU was powered from a lower voltage (ldo voltage) than the UDPI signal it would cause issues (maybe a stupid question…).

FWIW, when making one of my programming keys, I accidentally put the resistor inline with the VCC line instead of the UPDI line. If I remember right, the computer didn’t even recognize that the MCU was connected. So I just connect the pins directly at the MCU, without anything like LDOs or diodes in the path.