102,400 one-shots

Just between us, what did that cost?

I'd like to do some custom mixed-signal chips. Apparently it's not a totally crazy idea any more.

A mutual friend wants me to do another driver for his electro-optical gadget. Fast driver. How does one get rid of 60 watts on maybe half a square inch of PCB? And not block the light?

Charging and discharging capacitors, in principle, takes no power.

San Francisco is crawling with driverless cars that are decorated with spinning gadgets.

google waymo lidar

Somebody showed us a single-chip lidar rangefinder that works at a few inches and costs a couple of dollars. Can't remember who.

...

So, it's similar to a dynamic RAM versus static RAM storage?

If you charge 'em with a switch to +V, and discharge 'em with a switch to GND, it certainly DOES take power. Switches aren't zero-impedance, so it warms them (and/or the capacitors aren't zero-ESR, and it warms them instead).

Am 07.04.23 um 22:23 schrieb John Larkin:

Our wafer testers got rid of 50000 Watts for 4K test pins in less than a cubic meter, and it went this way: "My uncle works at DaimlerBenz/Mercedes here in Böblingen, I'll ask him" and he made it work. Masses of nekkid square centimeter chips, Alu block, copper tubes, cold water...

:-) Gerhard

If that were true, CMOS devices could be built that would operate at MHz, but draw no dynamic power. CMOS power dissipation is purely in the charge being shuttled when toggling the signal levels. I've never heard of any way around it, other than not changing signal levels.

There are some "pin driver" chips used in test systems, but as far as I can tell they are not sold to civilians.

It was surprisingly reasonable--even a dedicated engineering run (20ish wafers) was about $50k, I think, and the multiproject wafer approach (a la MOSIS) is considerably cheaper than that. This was on an XFAB 180 nm process with APDs and SPADs.

We've been kicking around doing the guts of a laser noise canceller on a transistor array chip--onsemi will do that for just a few $k.

A loop of copper tube with water running through it, or maybe a vapor chamber heat spreader with a finned heat sink someplace nearby. Vapor chambers are about 10x better than copper, iirc. Digikey sells them.

The canonical method for charging and discharging caps fast and efficiently is to use an inductor and diode to force the charge waveform to be half a cycle of a sine wave. That's how a lot of lidar laser drivers work.

Cheers

Phil Hobbs

The inductor + diode trick does it.

It's tough to get big enough on-chip inductors to do it with logic, of course, but maybe not absolutely impossible--kinetic inductance in small superconducting circuits can be startlingly large.

Cheers

Phil Hobbs

My Pockels Cell driver charges the capacitive load in a few nanoseconds, and then recovers the energy in a few more. That's why it is so small and runs off a wall wart.

It does use an inductor, so the architecture would probably be hard to replicate in an IC.

In theory, one can run up a parallel LC circuit and, once pumped, get a zillion cap charge and discharge cycles for free after that.

Then, at any zero crossing, short it out for as long as you like, and turn it loose any time after that to resume the oscillation.

Nice example of a differential equation with initial conditions.

As an aside, when I was at IBM we had a research group there, which among non-German speakers was usually called "Bubblegum." ;)

Cheers

Phil Hobbs

That's basically the dual of the inductor + diode thing, which switches automatically at the voltage peak.

+20V 0--LLLL-->|--*---------* 40V peak | | | LD Triple-stack Osram laser few nF CCC | CCC 0 | \ EPC GaN FET | 0 | | GND GND

The Uber one I took apart a few years ago had a total inductance (around that cap - laser - switch loop) of just under 400 pH, about half in the source and half in the drain. (Ascertained by dorking LTspice till the waveforms agreed with my data.)

Cheers

Phil Hobbs

I should add that Uber wasn't using the inductor trick--they had a 40V power supply and a resistor. Demonstrating that they weren't the same circuit was the point of the exercise.

Cheers

Phil Hobbs

Böblingen was the old HP campus, the wafertester was the Verigy V93K, ex HP. I had one reserved for me for 4 days a week. The tester chips had an onchip Mips 4000, a CPU I do like very much. There was an engineering hall with 300 engineers,

285 of which were software people. Even me, I integrated TDRs into a 10 Meg lines software system. Don't try to explain what a TDR does to a pure software guy.

A textbook example of how you can turn one of the world's best technology companies into a pile of debris ready to be eaten by other companies.

The IBM campus is 5-10 Km more south; a friend of mine was developing caches for IBM mainframes there.

Cheers, Gerhard

What do Germans call bubblegum? I'd expect a very long word.

I have two wonderful books.

The HP Way by David Packard 1995

and

The Journey by Carla Fiorina 2000

Everybody should read them. The second one will only take a few minutes.

Jim Thompson had an elegant CD ignition circuit that worked something like that. It was crazy simple compared to what I used to do on motorcycles: dc/dc converters and oil caps and SCRs and stuff.

My Pockels Cell driver charges an inductor, dumps it into the capacitive load, returns the energy to the inductor, and then dumps that energy back into the power supply. The big problem was inductor heating; that took three revs to get right.

Except for being crazy fragile BGAs, the EPC parts are magic. I think they are selling a packaged part now, which they swore they would never do.

Kaugummi.

lørdag den 8. april 2023 kl. 16.39.18 UTC+2 skrev John Larkin:

clever maybe, but needing a 5mH 5A inductor doesn't seem that elegant