Hydra v2p8 in test!


I am announcing that I have the Hydra 2.0 chip on my desk along with the test board and am currently subjecting the four community bandgap designs to characterization measurements.

If you want to know more about the test board I’m using, the project is in the IP catalog under “Systems”, catalog entry “Hydra Test”. That’s mostly just the printed circuit board, though. The circuit board has a USB port and an FTDI USB-to-serial converter chip, so that the chip can be accessed by software through a host computer. The software is called “tclftdi” and is on opencircuitdesign.com, but the web page (opencircuitdesign.com/tclftdi) is not advertised from the top-level web page, and the distribution is a bit out of date. I’ll try to get this updated soon. The current version of the software supports two different USB devices, both FTDI chip based, one of which is a hydra-like test board with the FTDI chip driving an SPI or I2C interface with bit-bang controls, and the other which is a Prologix USB-to-GPIB converter (this is the cheap alternative to the National Instruments device). This software allows me to write Tcl scripts that can set up the test chip through the SPI interface, set up the power supplies through GPIB, and then make current and voltage measurements, again through GPIB. My main instruments, all purchased second-hand through eBay, are an HP 6624A quad DC power supply, an HP 3478A multimeter, and an HP 5410A DCO.

If you want to know why it’s Hydra “v2p8”, that stands for “version 2.8”, and the “.8” comes mostly from iterations of generating GDS data, NOT iterations of fabricated chips! There were just two fabricated chips.
The first one was “mostly” functional, except that an unfortunate error in the PDK (caused by attempting to hand-edit foundry data, way back when both the PDK and the tools to support the PDK were being developed at the same time) mislabeled pins on the bipolar transistor and caused the three bipolar-based bandgaps to be nonfunctional. Since the whole purpose of the Hydra chip was to validate the community-designed bandgap circuits, a respin was mandatory. But the PDK error was the only malfunction, the respin was quick, and the chip I have on my desk appears to be fully functional.

The task now is to compare the measured data (which will be at room temperature only unless I want to try playing around with a cooler and dry ice, or a toaster oven, both of which I’ve done in the past) to the simulated results, so that we can validate the design tools and the simulation models.



First round of testing of 25 packaged parts correlates very well with simulated results. In particular, I measured the bandgap output over trim and applied voltage for each of the four community-designed bandgap circuits. I compared the mean and standard deviation of the results with the monte carlo simulations from ngspice. Obviously, 25 parts does not make for a particularly good statistical distribution, but mean values were well within the 1/sqrt(N) expected margin of error, and standard deviations were close. The ranking of the bandgaps from tightest output distribution over samples to loosest was the same for simulated results as for measured results. For me, the take-away message is that the ngspice monte carlo methods (and the X-Fab monte carlo device models) are accurate and should be a critical component of any analog design. Because we are using this testing to validate the methods, we did not use the monte carlo simulations as part of the challenge spec (which used simulations at PVT corners instead), but having validated the method, monte carlo simulations will be a part of any future analog design challenge.