How to miniaturize circuits?

[nitish111]

Hi all,

I was assigned to design a device with design inputs and I made a working prototype according to specs. I presented my fully functional design (on an enclosure of size 70mm x 128 mm x 25 mm) to my supervisor.Then he showed me a competitor's similar product which was about the size of regular 9V battery and said we want that device to be about the size of our competitor. Then I started researching miniaturizing circuits (I don't even know if it is called that). I don't even know where to begin with. I have 2 years experience is analog and digital circuits but have not done something like this before. I can make my components sizes smaller but still I don't know if it is going to work. I have heard of custom ASIC but I don't know how it works. If I have a schematics, do I just submit it to these microelectronics fabrication company? Where do I look for information? Any help would be highly appreciated. Thanks!

[suicidaleggroll]

More layers on the PCB, intelligent component placement, and if necessary blind/buried vias will allow you to get the board size down to approximately the total surface area of your components.  In combination with that, using smaller components where possible will let you get things quite small.  If it's still too big, then you can look into custom ICs, stacking, etc.  It's really application-specific.

[Benta]

Before going further, look at the business case. How many are you going to make?

Analog and mixed-signal ASIC design is a big mouthful, with NRE costs running into millions of dollars (as opposed to digital ASICs, where gate arrays are reasonable cost).

If you decide to go ahead, expect that most of your time will be spent at the semiconductor design lab.

Mostly, only the 800 lb gorillas in the electronics business have the dollars and resources to go into this.

edit: suicidaleggroll's advice is sound, that's the way I'd myself.

[Siwastaja]

Can you post more details? For example, what are the biggest components right now?

Very small packages - including chip-level packaging (i.e., no package at all, just the semiconductor die with solder balls directly soldered on PCB) - have become commonplace in the last few years.

Custom ASIC is most likely simply "no"; this is basically one of those cases where if it was possible for you, you wouldn't be asking . After you are selling hundreds of thousands of units a year, take a look at it again. Although there always are some rare corner cases where it makes sense and is possible even for smaller scales, but they are so special I'm taking chances to just say "no".

[Fire Doger]

I have heard of custom ASIC but I don't know how it works.

Application specific Integrated circuit. It's like you design your own IC. If your design has a mosfet, an ADC and a transistor then you have to design them inside the same silicon die. According to your questions it isn't something that you or your company can do.
The smallest you can do is "chip on board"
If you have competitors product available look how they did it and use the same techniques. if they use 0201 resistor don't use 0805 in yours...
It will work just fine if you don't make any mistakes on power dissipation. For signal integrity you can use multiple layers as suicida said.

[Ice-Tea]

Don't reinvent the wheel. You can make a fully discrete DC/DC controller with 5 opamps, opscillators, integrated linear supply,... Or you can use a SOT-5 that does the job for you. Designing a custom IC is probably a no-no but chances are you can find one that combines a lot of functionality already...

[nitish111]

The device is a pulsed current source that can sink upto +-20 mA upto load of 5-7 kohms. So it has two DC to DC converter to generate +100V  and -100V. I used the datasheet recommended components so there were 3 big inductors ( 6.8×6.8×3.0 mm), multiple diodes (SOD-123-2), 1 DPAK regulator and 2 DPAK power transistors, 5 MSOP-8 ICs, 48 pin (LQFP) MCU, crystal,2 (2220) capacitors for high voltage bypass,  0805 capacitors and resistors. The company is small and project budget beyond $ 2-3k is out of scope. Mostly I have to work with off the shelf components. I could use smaller components but I would have to handsolder them for a working prototype.

[Siwastaja]

OK, so you have DC/DC power of about 2W. For that power level, a 7x7x3mm inductor is a bit big. Try to work on that, although, I think you can save this as the last step since you have easier optimizations first:

DPAK power transistors are definitely much bigger than necessary - power MOSFET cases have been shrinking a lot lately. I'm switching hundreds of watts of power with transistors half of that surface area! So I think this is #1 where you can miniaturize, since this is not even much power at all.

DPAK regulator - I guess this is for some logic and/or transistor drive. Probably not much more than a few mA? Definitely use a smaller part.

Handsoldering is not a problem whatsoever in these things. You can even solder BGAs, QFNs etc. with a $50 hot air station quite well.

0805 are HUGE. I'm handsoldering 0402 passives all the time. Even though a little bit more tedious to handle than 0603, 0402 is a good sweet spot for easy enough prototyping and good manufacturing yield even in lowest cost assembly houses - and good integration. 0402 tends to easily fit to bypass 0.5mm pitch IC power pins without requiring huge fan out area that gets in the way of everything else, like a 0603 let alone 0805 does.

Are the 2220 capacitors ceramic? If yes, they are super expensive and crack easily, destroying the device. Replace them with some 1210 or smaller. Use several in parallel.

Stacking multiple PCBs might be an answer, but you have a lot of room for improvement before that.

Just take the bullet of handsoldering these things. You'll learn while doing, and it's not going to take long. Expect some unreliable prototypes you need to rework a bit. But aim for commercial manufacturing as soon as possible so that you don't need to do massive number of prototypes.

Basically, do the parametric search on Digikey / whatever you use, and choose the smallest size you can find, for each component. Try to avoid cases where the pins are "under the device" if you are unsure about hand prototyping. OTOH, those are often the smallest, so you may need them.

Do you really need a crystal? Is timing so critical?

[Neomys Sapiens]

The coils and the DPaks look indeed a bit large for your requirement.
What the previous posters mentioned as 'stacking' (splitting the circuit onto more than one PCB) is best done by taking a step back first. You might even have to spend some (area, volume..) to be able to achieve this, because you might have to redo some internal signals to be suitable for a less close coupling.
It should be obvious from the manufacturer's layout recommendations that you cannot do a split right in the middle of a DC/DC-converter when you run out of board space. So look for a sensible modularity first.
Do you need a 48-pin MCU or could you go for less signal lines if you convert parallel to serial interfaces or busses?

[jbb]

100V 20mA could just kill someone (they’d have to be unlucky), so don’t neglect safety. Safety of the completed product and safety for you while you work on it.

Going to filled via PCBs allows via-in-pad designs, which costs more money but helps pack components in. If you’re looking a fancy PCBs (blind/buried via or via in pad) you should talk to your PCB fabricator now to see what makes sense; there are big price changes for seemingly small technology changes. I suggest you also consider how these will be tested and depanlised (v score, mouse bites etc.) without damaging nearby components.

Double sided loading (I.e components top & bottom) can be a big help (especially if using blind bias), but you need to be careful, because it means 2 trips through the oven. Try to keep all the big components on one side of the board.

Connectors are always tricky for this sort of thing. It will be tempting to just solder wires on, but that’s notoriously unreliable.

As Blueskull said, a dual output flyback may be better than two separate DC/DC converters.

You also need to look at your output spec.
How much ripple is OK?
How tightly regulated does it need to be?
What about short circuits?
Is overshoot OK?
How fast do the rise & fall times need to be?

[jbb]

The LT8301 flyback controller could be useful. It’s highly integrated and doesn’t need an optocoupler for feedback.

Also, does the output signal use the same ground as the input?

[nitish111]

Thank you all.

This is a lot of information to chew upon. I will definitely reduce the size of passives to 0402 or 0603 according to power requirement. I had 48 pin STM32 MCU which I can replace with 20 pin TSSOP. I will find a replacement for power transistors with low profile. I was using LT3580 as DC/DC converter but I think I will use a flyback controller instead. I will be working on a flyback controller for the first time so it there is any literature to begin with I would highly appreciate. Any information on low profile transformer as well. LT8301 should be a good datasheet to begin with. The output has to be regulated with ripple less than 100 mV. I need pulse width of minimum of 5 us, that is why I had external crystal. I will probably have 2 board construction once I finalize enclosure. Output signal has same ground. 

[jbb]

[qoute]100mV ripple[/quote]

I thought you wanted a +-20mA current output? Also, what accuracy and stability do you need? Do you need a fixed output or a variable output?

[qoute]minimum of 5us[/quote]

That’s quite short. I expect you’ll need to have a 120V (?) DC rail and then some kind of output switch/regulator to produce the output. Does your current prototype achieve this?

[ruairi]

I was assigned to design a device with design inputs and I made a working prototype according to specs. I presented my fully functional design (on an enclosure of size 70mm x 128 mm x 25 mm) to my supervisor.Then he showed me a competitor's similar product which was about the size of regular 9V battery and said we want that device to be about the size of our competitor.

I might be stating the obvious but how did your competitor do it?  I'd be cracking the lid on that unit and learning as much as possible.

[Dubbie]

I had 48 pin STM32 MCU which I can replace with 20 pin TSSOP.

Why not go to QFN package?. They are relatively easy to hand solder with hot air and are nice and small.

[phil from seattle]

Do you understand why the competitive product is so small? I'd tear it apart (or have some one else do it). I would want to know what corners they cut to get the size down. Are you over-designing for the market or are they under-designing? I'm not suggesting copying them but it helps to understand what you are up against.

[nitish111]

I have a DAC feeding bipolar voltage to howland current pump to generate +- 20 mA with +100 V and -100V rails. On a 1 k load there is a overshoot of about 1V with my current prototype. Output current is variable set by DAC. Only problem with this design is that DC/DC heats a little bit.

I asked for the competitor's device and they said that they got the device as a loan for couple of weeks and soon they might be able to get me one for a teardown.

[schmitt trigger]

In addition to everything else....
Flex boards sometimes  allow you to use otherwise unused spaces.

[jbb]

I have a DAC feeding bipolar voltage to howland current pump to generate +- 20 mA with +100 V and -100V rails... Output current is variable set by DAC. 

I'll bet that Howland pump is a little painful (in cost or size).  If you use a flyback converter to produce floating +-100V rails it might be possible to reference your regulation electronics to ground, as in attached image.

It might not offer the dynamic performance you need, though.  Some simulation (or, even better, analysis) would be required to tweak the response.  (The dynamics will be influenced a lot by Cx, MOSFET Crss parameter, Rg, Rfb and Cf.)  Additionally, there will be some small DC offset in the output current, which might cause Cx to charge up to +100V (or -100V) when no load is attached.