Once upon a time, I had a summer job as a parking-lot cashier. Concert goers would come to the venue and shell out four and then five dollars for a spot in an unlined dirt lot. Fans of the Grateful Dead would offer “alternative means” of payment or perhaps a bag of pennies. I’d send them on their way to take their chances without a parking voucher. We would get off work about the same time as the warm-up band went off stage. Then, we were free to watch the headliner. This went on for five years at Shoreline Amphitheatre.
I quit after Bill Graham’s unfortunate helicopter crash. The other cashiers were out of a job once Clear Channel took over and folded the parking pass into the ticket price. Only the larger acts could book back-to-back weekdays and expect to sell out. Janet Jackson and Bruce Springsteen come to mind. The biggest acts, Brit-bands like The Who?, The Stones and Pink Floyd would book stadium venues and we got in on that in various ways.
So, anyway, Springsteen came through for a Tuesday/Wednesday set. I worked them both but was tired and hungry after the second day. I got my employee discount slip and stood in the employee line for a hot dog and then made my way out the nearest exit which happened to be the main entrance. People lined up to get in were incredulous.
“Are you Leaving?”
“Yeah, I saw him yesterday.”
“You saw him yesterday, and you’re LEAVING?!?”
Image Credit: Author - Just for reference, a view from the General Admission lawn seating at Shoreline
Bruce was at the top of his popularity and they just couldn’t fathom that someone could be jaded enough to want to rest up for work the next morning instead of seeing “The Boss” again. This was the pivotal moment of their whole summer and it was my Wednesday night. Going out on an interview is like that. It’s a neurotic, exciting, and ulta-important day that the interviewee will remember in excruciating detail. To me, it was the day before Thursday at Google.
That the Googleplex is right across the street from the amphitheatre is just a coincidence. Another coincidental turn happened at Microsoft just up the street from there. A coworker asked me if I remembered her. I looked at her badge for a clue, came up empty and I had to answer, “No.” It turns out that I had interviewed her a few years before.
In a way, she was like that Bruce Springsteen fan, a face in the crowd. What made people blend in is that I was asking the same questions of everyone. If it was a pretty girl, that was nice but the answers were the answers. Reading the resume, printing my questions, doing the interview and writing the feedback meant that I needed to jump back in and make some progress on the design of the moment. I went over my notes on the interview and gave my assessment as objectively as I could.
Like a teacher who grades every exam by the same rubric, the candidates funnelled themselves into their buckets by their answers. If you’re ever faced with a hiring spree and selected to find the best candidate(s), then maybe questions like these will help get you there.
Image Credit: Author - Googlers! Left to right, David, Choon, Ken, Lily, me, and another David. I was notorious for organizing lunches involving the bike built for seven. The man behind Choon is Ajay. It’s like six big brains and me.
Before I start, let me say this. What I don't like are open ended questions. "Tell me about high speed. Describe the 'flow' of a layout. How do you collaborate? Why do you want to work here? Anything where the answer can wander ends up being a rabbit hole. Instead, I give them the "magic word" instead of having them guess at what I want them to say. Let’s get after it.
1. Describe the pad stack you'd use for a 0.4 mm BGA.
To get them started, I’d let ‘em know that I could fill in any details about the product or the process that they would find in a datasheet. This lets me know if they can do library, know IPC standards, break IPC standards, understand via-in-pad, among other things. In a way, it's a survey question so I know how others in the industry are approaching the problem.
A ten-second answer can carry just as much weight as a full five minute discussion. While I already said that I’m not looking for magic words, there is a vocabulary that goes with this question. If they ask about ball-diameter or if I want mask defined or non solder mask defined, collapsing or non-collapsing, we may have a winner. If they volunteer the metal, mask and paste, we have at least a B-player.
Below that are people who are winging it. For the wingers: Mid-answer, I'll tell them that I want the three values. I'll let them work in mils if they prefer. (not a good sign since all the data sheets are metric) I'll tell them to start with the notion that we want a solder mask dam between the pads and to start with the IPC minimum. We want some offset from pad size to mask size, we want the via-in-pad and so on hinting at the final numbers.
Again, a good answer can be one that they know off-hand but it's fine if they have to work it out. Generating a footprint for a 0.5 mm pitch device can be done with typical IPC standards while 0.4 pitch requires more precision, maybe even lasers! You see, a simple question can reveal a lot. I’m going to gloss over the analysis for the rest of these questions though the depth of what I’m trying to learn in the short time we have is just as nuanced.
2. Let's build a 4-pin BGA on the whiteboard.
Again, deceptively simple. We have a few things to capture beyond a square with four circles. Silkscreen size and minimum stroke thickness, typical component height, courtyard area for clearance and/or underfill, DFA rules, orientation marking etc. This is only a preparation for the next question.
3. Let's turn this into an SMPS circuit.
Now we're imagining a big inductor first, caps on In and Out a Ground pin and a fourth pin for control/feedback/current sense, whatever they come up with as the essential ingredient. If they can sketch out the schematic symbol and add the passives, so much the better but I will do it and knock them down to a B.
Knowing how these little parts function in comparison to a linear regulator used to be a standard question. I dropped it from my list as it was dragging down the average scores. We can discuss the connection on the output side where the switching noise lives. How many ground vias, copper weight and other practical aspects of PCB Design were brought up instead.
4. What if we want this to be a 20 pin 5x5 BGA with four power supplies?
This is a PMIC test and it gets more complicated managing the huge inductors and the four extra pins that do "something" from those "essential ingredients" plus external clocking and other goodies. If someone is struggling with this, they may have a hard time with the rest of the story. That this comes up again only stresses how important power is in a mobile device.
5. We're ready to dive into a software question.
Constraints drive the particulars. There are a lot of constraints to cover and we're probably about half way through our time slot. It’s best to cut to the chase. How do you create a match group or control phase tuning limits? Those are better questions than letting them grope around the editor. Knowing the software is important and the constraint manager is the bellwether of design knowledge.
Image Credit: Author - A slice of one page of one tab of one set of a sprawling constraint manager.
We continued with number 6. USB-type C connectors
A discussion of what to do when the data sheet tolerances exceed the fab-vendor capability or how to approach the fan-out are instructive. This is a stand-in for all of the odd parts on a board.
7. ESD diodes are the next link in the chain.
This is a chance to explore the uncoupled length of the superspeed differential pairs and, of course, my favorite net - the Gee En Dee!
8. Transmission lines are the crucial conduit across the board.
PCB stack-ups and materials are in play along with the various trace geomeries that makeup the art of RF and high-speed digital design.
9. SOCs, the “boss battle” of PCB layout.
There’s no getting around these beasts. If the interview flew by, there is enough depth here to fill the time. There is a hierarchy of signal groups to discuss. If the pace was slower then we’ll skim over the highlights of decoupling caps and thermal management.
10. What Goes Into a Document Package?
Surprisingly, not every designer does their own tape-outs. We did. In some cases, we were responsible for procurement as well. Managing the vendor whether it was a local fab shop or a far-away ODM (original design manufacturer) required good presentation skills and an eye for detail. The principle of documenting everything once and only once is an imperative.
11. (If Time Allows) Tying Number Nine Back to the PMIC.
When an interview goes quickly, we come full-circle to the power plant of the PCB. All along the way, there is as much inferred as said. For a librarian role, about half of these questions would suffice but we’d go into more detail on a nuts-and-bolts level. This list was developed over time but the process of taking each question as a determining factor was the basis of my system. By the end of the interview, it was pretty clear if they were able to check the boxes of the job description.
Finally, we end with Q&A that allows the candidate to drive the last ten minutes or so. If we started with one hundred points and took them away for partial or incorrect answers, this is a chance for bonus points to be added back on. Using a grading curve to get a level playing field and reduce judgment calls. As many as 10 points, the total possible for each of the above, could be given back for interesting questions.
I went a long way just to say this: Finding a rock-star among the faces in the crowd isn’t always easy. Direct questions with one central answer works better for me than handing the mic over to someone and cutting them loose to do “whatever”.