RE: [SI-LIST] : To overshoot or undershoot: ( to diode or not to diode)

Mellitz, Richard ([email protected])
Mon, 23 Feb 1998 13:38:57 -0500

Andy,

Yup I'm taking about current in the clamp diodes doing bad things! I
have experienced large differences between parts, but spice seems to
predicted this. I guess a vendor provided max current into the diodes
would help.

Many PCI designs I've seen clamp heavily on falling edge. I think the
PCI clamp curves are spec'ed. Is the max current spec'ed in PCI?

... Rich

>----------
>From: Andrew Ingraham[SMTP:[email protected]]
>Sent: Monday, February 23, 1998 9:09 AM
>To: Mellitz, Richard; '[email protected]'
>Subject: RE: [SI-LIST] : To overshoot or undershoot: ( to diode or not to
>diode)
>
>I take it you are talking about the currents into or out of clamp diodes
>that is doing bad things.
>
>Some ICs specify a limit on the currents into those clamp diodes. Have
>you seen problems where this current limit was not exceeded?
>
>I believe the mechanisms for overshoot and undershoot clamps are totally
>different (but I am not an expert on this). Undershoot clamp current
>goes into the substrate where it affects neighboring transistors, and
>has the potential to activate those parasitic SCRs that causes latch-up
>and total chip failure if stressed enough. Overshoot seems to be
>different.
>
>I have been told of a situation where some sort of programmable IC, was
>seen going into program mode, on account of overshoot voltage magnitude.
>These ICs typically use something like +12V applied to one or two pins,
>to tell them to go into program mode. The overshoot on some pin went to
>something like +8V, which was just enough. I think it wasn't even the
>pin you're supposed to use to activate program mode (but I'm not sure
>about that).
>
>Sometimes there are huge differences in the clamp diode characteristics
>even between the "same" components from different vendors. We recently
>looked at one such part, a simple logic IC. SPICE models from the two
>vendors showed an enormous difference; on one model they were brick
>walls, on the other, there was significant resistance. So we measured
>parts from both vendors, and indeed, they were different. Same generic
>part number.
>
>We have had situations with RAM chips (different part numbers) where one
>vendor's product had no overshoot clamps, and exhibited significant
>ringing and timing problems due to ringback.
>
>Regards,
>Andy
>
>