Zero FIXMEs--BEOS 3rd Ed is about ready to send in

On 01/12/2020 14:04, Phil Hobbs wrote:




I think you may have just suffered from the Osborne effect, as I will  
now hold off my purchase. The only way you can redeem this situation is  
to tell me about all of the excellent content you had to remove from the  
new edition, so that then I have to buy both.

Oh, and I've just missed the deadline to tape out a high-bandwidth  
digital pot on skywater's 0.13um CMOS. Installing the open source tools  
took me 90% of the available time. I then found the learning curve for  
MAGIC very steep indeed, and the user interface quite medieval, (and I  
am giving it the benefit of a lot of doubt, due to my accumulated hatred  
of Cadence). I think KLayout might be more usable in the long run,  
though I hear that the DRC rules and the generators for the device  
layouts still need doing for that one. For schematics, xschem is very  
nice, though there is some intermittent bug (in debian?) that prevents  
it from getting correct mouse coordinates when a keyboard event happens,  
when run in a qemu VM on my machine. Maybe I will get these things  
sorted out in time for their next free shuttle, though by then more  
other people will have figured out the tools, so it might be oversubscribed.

On 12/1/20 7:30 AM, Chris Jones wrote:




Oh, darn, another $10 down the drain. ;)



How fast was it going to be?  The quickest one I know of is the 1-k  
version of the AD5273, which is about 6 MHz at half-scale.  You can get  
MDACs up to 10-12 MHz, e.g. AD5432, AD5452, and DAC881, but that's all  
she wrote.

Cheers

Phil Hobbs

On 02/12/2020 05:52, Phil Hobbs wrote:







I'll have a go at simulating it, though I don't yet trust the models. It  
might take me a few days or more to get around to it. I'm guessing  
without trying too hard, it would be in the low hundreds of MHz.

If you have *lots* of spare time, you can also have a go, as there is a  
lot of opportunity to improve performance by changing the topology to  
suit your specific needs. I'd suggest using only NMOS switches, since  
the wiper could stay near ground in your cloud nine circuit. It might be  
worth figuring out some trick to use fewer switches/taps on the resistor  
string than the resolution would imply, e.g. by somehow dorking a  
resistor at one end of the string, or finding a way to make each switch  
not have to be big enough to deal with the total current, e.g. by  
feeding the current into several taps simultaneously. Also, using some  
sort of force / sense arrangement with two wipers may help.

Here are links to information:

https://groups.google.com/g/skywater-pdk-announce/c/75Mt_qw4WBA

Here are some instructions/scripts for installing tools (including some  
you won't need/want) (tested with Ubuntu 20.04, I gave the VM a 60GB  
disk and it needed much of that):

https://github.com/yrrapt/sky130_setup

https://github.com/bluecmd/learn-sky130/blob/main/schematic/xschem/getting-started.md

On Tue, 1 Dec 2020 13:52:32 -0500, Phil Hobbs





We use analog multipliers to simulate things like eddy-current blade
tip sensors and oil debris sensors.

We use DAC8812 in a capacitor simulator, serial data and 10 MHz ref
bandwidth. I think most MDACs have different bandwidth for different
bits.