Where should you put the DCDC's that supply DSP IC's etc?

[Faringdon]

Recently i  failed an interview  because of  not recognising a 4 channel interleaved Buck converter controller IC which was supplying Vcc to  two DSP IC's. ....
....So , i was shown a photo of a PCB..... i could see the two big 100+ pin DSP IC's (they were in fact pointed out to me). They were some 5cm apart. Right In between them, was an approx 24pin square IC. The question was..."what is the 24pin IC?".
I looked around the IC, and could see no inductors, and no power FETs or diodes, so i did not think it was an SMPS controller. So i said it could be a microcontroller......but no...wrong....it was in fact a 4 channel interleaved Buck controller IC, supplying Vcc to the DSP IC's. ....Interview immediately failed.

On closer inspection, i could then make out that the  "large ceramic capacitors" (approx 2512 size) which i thought i saw, were in fact, SMD inductors.......i should have realised that being grey instead of brown, they were inductors, but to be honest, the shading on the photo wasnt that distinct.
The thing is, since they were using those SMD type inductors, it indicates that the current throughput couldnt have been that high. -As such, why did they have an SMPS right next to the two DSP IC's? Shouldn't they have had the SMPS more distant from the DSP IC's, (due  to potential noise issues) and brought the current in on wide tracks? I mean, it cant have been that much current if they were using those SMD inductors (the ones that look the same shape as 2512 smd capacitors).
They were making out that it should have been obvious to me that two large DSP IC's would need an SMPS right next to them......when in fact, surely  having an SMPS right next to two DSP IC's is a "last resort" situation. Would you agree? Or am i wrong, and is it "par for the course" to put SMPS's right next to big DSP IC's?

[georges80]

Excellent, the interviewer obviously weeded out the interviewee rapidly and the interviewee could not waste more of the interviewer's time. I'd suggest that the company has a good selection process, probably a good place to work at 

The rest of your post confirms the interviewer made a good decision in the case of the interviewee.

cheers,
george.

[SiliconWizard]

Uh, is any of you two serious?

[commie]

I take it, they didn't offer to pay your traveling expenses either?

[Siwastaja]

Low-voltage (say, from 5V to 1.2V) DC/DC running at high f_sw (like 2MHz!) can use such small inductance value (a few uH) that even a tiny 1210 size inductor can have saturation and RMS current ratings of many amperes. Having four of those, this could easily be a 10A converter. Yes, modern switchers are tiny! This could easily suffice for two high-end DSPs.

There is nothing wrong placing the switcher between two DSPs. If the layout is decent, and it has to be at high f_sw, high dI/dt is contained within tiny area. High dV/dt node is also small, might be just a few square millimeters, given that tiny inductor.

Granted, maybe if you do high precision analog, maybe you want to give another inch of physical separation, but then again, the high precision ADC would be elsewhere, the DSP itself is fine with some noise.

Regarding the question in the title, the answer is, because modern CPUs run at very low core voltages at high currents, the power converters have to be close to the CPUs. Really due to basic Ohm's law.

Your problem seems to be getting fixated into the process - who's right? is this right? - instead of just taking the opportunity to learn from your failures and go on with your life so that next time you understand how tiny power inductors really are today, and you can even copycat that design idea yourself, take advantage of it instead of questioning it for no good reason!

[SiliconWizard]

UH yeah and keep in mind that the main reason you "failed" the interview may not have been that you failed to find what the damn IC was. Interviewers (decent ones anyway) do not expect people to have an exact answer for everything. Rather, they are interested in hearing your thought process, which would elicit how you reason, what's your experience, etc. If you got fixated on giving the "right" answer, possibly that made you look like a typical "beginner". That's how a beginner would react to a question they have no direct answer for.

You said you realized - probably after the interviewer told you what the IC was - that what you thought were capacitors were in fact inductors. But did you say, while you were trying to answer (and before the interviewer told you what it was), that you thought they were capacitors, for instance? That may have helped them see what you had in mind, maybe the would have hinted those may not be capacitors, which could have led you to the right conclusion... etc. See the point? (And with that said, I wonder how you could confuse ceramic caps with inductors.)

But judging from a photo rather than a real board you can manipulate, yeah, that would have made things harder. That was possibly the point.

Oh, and maybe there are also other parts in the interview that you "failed" without necessarily realizing it.

[Siwastaja]

Some 1210 inductors really look like MLCCs, especialy if it's just a photo and not an actual board, so I'm not surprised about that. OTOH, if you have seen enough modern designs (by others, or designed them yourself), you know what to expect. That "a bit different MLCC" among banks of MLCCs, near a mystery IC, has to be the inductor.

[T3sl4co1l]

Or even fully integrated parts, which are starting to make inroads.  And I don't just mean molded combo parts like LT uModules.

Saw one a couple years ago, some little Intel thing (I forget the brand, another one they've snatched up in recent years) that runs at like 10MHz, doing a few amperes in a QFN.  It went into some stupid brownout mode when the least bit of overvoltage was applied to the feedback node, or at least as far as I could tell that's what it was.

Being a DSP doesn't mean much to me, either; it could be a little MCU thing that just sips power and uses any random regulator, it could be a beefy boi that needs heatsinks and you can't even see it under there.

Tim

[Faringdon]

because modern CPUs run at very low core voltages at high currents, the power converters have to be close to the CPUs. Really due to basic Ohm's law.

Thanks, yes, thats the only reason you would expect the power converter to be right nect to the DSP chips (as well as lack of space anywhere else)....but since there were no noticeably bulky power inductors, and no fets or diodes in sight, it really didnt seem like a dead ringer for a high current 4 channel buck controller....and even if it was 10A....that could have been brought in on a wide enout/thick enough track from a more distant location.

It looked like a 24 pin micro. But the way he told it, was like "obviously the supplying SMPS is going to be right next to the DSPs"

[SiliconWizard]

because modern CPUs run at very low core voltages at high currents, the power converters have to be close to the CPUs. Really due to basic Ohm's law.

Thanks, yes, thats the only reason you would expect the power converter to be right nect to the DSP chips (as well as lack of space anywhere else)....but since there were no noticeably bulky power inductors, and no fets or diodes in sight, it really didnt seem like a dead ringer for a high current 4 channel buck controller....and even if it was 10A....that could have been brought in on a wide enout/thick enough track from a more distant location.

It looked like a 24 pin micro. But the way he told it, was like "obviously the supplying SMPS is going to be right next to the DSPs"

It's certainly absolutely obvious to someone that knows the board. If the guy really told you this, he's an idiot. Either that, or you made him angry for some other reason.

If you don't know what the DSPs are, how much current they draw, what their required voltage rails are, there is absolutely nothing obvious about this. Did you have this information beforehand? As if all DSPs in the world were strictly identical...
And, said DSPs may actually have some critical analog I/Os which could make placing a 'big' SMPS right next to them unpractical.

So yeah, we don't have the whole conversation, but from the little you said, the guy sounds like a smartass. But again, maybe you said, or did other things during the interview that he did not like, and this event was just a pretext to cut it off.

[commie]

All I can say is congratulations, it they're that stupid that it came down to a rudely interpreted / handled "name that tune" blurry ambiguous photo recognition contest then your interviewer is probably an arrogant jerk you wouldn't want to be your colleague / manager / ... anyway, so you win by NOT working there.

I agree 100%, but this is the 'system' that we find ourselves in, particularly here in the U.K.

[Silenos]

The thing is, you were not expected to tell the pointed ic in 30 s, you were to present the thought process of figuring it out... well, and boast with experience and acumen. 
Imo the SMPS was just a random point. Say, you got the job, and you are tossed a pcb from competition product with task "figure it out and make ours better and cheaper".

But anyway, just my thoughts:
- considering a joke with a smile: right away "I have no idea what it is, I'd just type the package marking into browser to get over it". - but that's actually on point as it is usually an initial step to recognize an unknown ic.
- "oooh is that an AD ic? they noice but pricey..." - assuming you can recognize different producents logotypes, and have some comment on their products coming from an experience.
- "can I have the photo of opposite side of the board? Or just give me the actual board, I can't see shit on this photo, my eyesight fails me or sth... Maybe there are additional parts for the IC, like frequency source for clocked ic, or heatsink stiching from the exposed pad of the ic, as QFNs can have exposed pads, and power ic may require either external heatsink or some pcb-made, though things like fast interfaces transceivers/phy also can have heatisinks...nope, no ethernet or usb on board..."
- "I didn't had a look around the whole board, so I can't tell where the DSPs are powered from so maybe that's the power ic... are those DSP require multiple voltage levels and full amps of current? Big cpus, or fpgas usually reqire so... the bigger and faster ic, the more amps it sucks, and the voltage spec are tighter and tigher... Those DSPs are from.. TI? don't know, never worked with TI DSPs"... and so on.
- more on that ic: smps controllers have strict routing rules, producer usually draws that in the datasheet or app notes, you can usually have bloated copper around inductors and flyback capacitors, maybe big fat tracks with multiple stitchings, sinking into inner layers (thats why bottom of board helps to tell it; and these can be visible, outside of IC - under QFN with exposed pad there is usually no space to route underneath). Multichannel ic have repeating patterns of such structures...

Or, say some C++ interview. Noone is expected to recitate the standard or know by heart some obscure syntax of edge condition template crap.
The correct answer is often sth like "I heard/read/saw about it, the issue is how compiler deals with X, it propably can be solved by doing Y (I'm not sure, but that's possible to check by myself,) with Z or sth header file, some syntax magic since iirc c++17, with some obscure "auto" special meaning introduced in another newer standard in last decade, and with some new keyword I forgot... but definitely it is explained in the reference of Y or Z. But hey,  that's is still a progress, as in the past there was no way to do this. That's a point for java  ".

[cgroen]

Maybe they already knew you from eevblog forum ?
Or, maybe there is more to the story ?
Do you agree......