https://open-source-silicon.dev logo
Join Slack
Powered by
I'm exploring chipignite and efabless as a solutio...
# chipignite
a
I'm exploring chipignite and efabless as a solution for some asic applications. What is the limit for bandwidth of input analog signals? I saw one page mention 50MHz, is that correct?
d
I was in a discussion the other day where this came up. It depends on how good a signal you want. It starts to attenuate after 50MHz but you can still detect switching up to close to 200MHz or so. There are ways you can design and use your own io cells, too. We taped out an open frame chip on the most recent shuttle where the customer implemented a separate IO ring inside the customer area.
👍 1
a
This is helpful, thank you!
t
@David Lindley What do you think is the biggest contributor to the limit ? Is it the ESD structure / difital input buffer / digital output buffer / something else ?
s
r u talking about me @David Lindley? who did
t
The limitation of 50MHz (output; 60MHz input, according to SkyWater spec) applies to the GPIO pads, and the limitation applies specifically to digital signals through the digital input and output buffers. However, the digital buffers create a lot of capacitive load on the pad, so analog signals don't have a lot of bandwidth above that. The "caravan" chip has unloaded analog pads which have been used in applications showing several GHz of bandwidth.
👍 2
t
@Tim Edwards But those in caravan have all the ESD protection removed as well right ? What I'd like to be able to do in the future for Tiny Tapeout is have some of the pads of the OpenFrame replaced by ones without the digital part and only leaving the ESD diodes in there. BTW, it's of course anecdotal, but yesterday I tested a differential receiver made in Tinytapeout receiving a 400 MHz clock through the default pads 😅
d
@Tim Edwards A couple of the analog pads in Caravan have the clamps pinned out separately, correct? They can be connected to the pad or not. As we start considering enhancements to our offerings I think a set of GP analog IOs added to openframe would be good. A single pad cell with no IO buffers and an input directly connected to the pad, an input with an ESD resistor inline, and maybe the big clamps pinned out separately, too.
t
@tnt: The 400MHz doesn't surprise me; it will be highly attenuated, but as long as the receiver is sufficiently sensitive it should work.
@David Lindley: Clamps are not ESD diodes. They are appropriate for power supplies but not for signals. I would like to do exactly what Sylvain suggests, which is to either update the existing caravan with most of the analog pads replaced with ones having ESD protection, or a 2nd version of caravan with ESD protection on the analog pads, or else a configuration file and script that allows each analog pad of caravan to be selected as having no ESD, minimal ESD, or full ESD. (The last one is difficult to do if you want to swap out digital pads for analog pads, but swapping out one type of analog pad for another type of analog pad is easy.)
👍 1
@David Lindley: I would also like this to be on our roadmap for a near-future product offering.
t
@Tim Edwards Why is switching digital to analog more complex ?
Ah, wait, it's caravan so there is some connections to the wrapper.
But in an Open Frame setting where the bare GPIO "block" is exposed, it wouldn't matter much ? Have a config file where we can choose for each position in the frame what type of IO we want, would be the dream for us 😅
t
@tnt: No, the problem is just that if you switch from an analog to a digital pad or vice versa, the connections into the user project wrapper change, so the user project wrapper pins change, and the XOR check changes, and everything becomes way more complicated.
@tnt: The openframe has its own user project wrapper; it's just that unlike caravel and caravan, it extends out to the padframe.
Properly constructed/automated/verified, it would be great to have complete flexibility to swap pads at will.
👌 1
t
Ah yes, I see the issues with the exposed pads. I guess you could maybe maintain them the same and be "NC" when they're not applicable to that pad type, but yeah, not elegant 😅
7 Views
Open in Slack
PreviousNext
Powered by