To respond to this other comment ("In custom analog land, you might as well forget all this.."):
This method may not be applicable to custom analog designs. These products typically have custom I/O cells such that the method would not work anyway (must stress a minimum of 30 identical clones). The pre-work required to validate the method, limits the practical application to only high pin count products. Thus the method was designed with a narrow applicability. However, there are already several large processor designs with several hundreds of cloned IO pins in production that can greatly benefit from this sampling method.
After much more scrutiny during the last 3 months a final optional test standard to be used for products with numerous cloned IO pins has been developed. This is expected to become part of the JEDEC/ESDA Joint HBM Standard in the near future.
Thank you again for your response and feedback!
We are sorry for the delay in our response as this was brought to our attention only recently, but we appreciate your comments. Here are our specific replies to your concerns.
"There are some assumptions (or ignored effects) here that bear checking. Particularly, that the I*R drops in the bussing are insignificant.."
This method was developed primarily for high pin count digital products. These products often use a common general purpose I/O cell that is cloned to form I/O banks. Many of these designs have distributed power clamps in each placement of the general purpose I/O cells and they are designed to have the same effective bus resistance to the supply clamp. Effectively these cloned I/O cells are identical. Once established through design checks (mandatory) that the pins are clones in every sense of the definition, the method uses random selection of these clones (a minimum of 30) to study their variations. The distribution of the data is then used to predict the unmeasured clones. All or any variations of bus resistance and inherent process variations are automatically covered under this approach. There are enough built-in checks for the method with very conservative criteria such that a 99% confidence level is guaranteed for application of the method.
There are some assumptions (or ignored effects)
here that bear checking. Particularly, that the
I*R drops in the bussing are insignificant (not
true on large "low power" chips); that the same
ESD clamp cell is guaranteed to be hooked up the
same everywhere it's used (you can get very
different results by feeding a stripe from
opposite ends, vs same end in/out) and of course
the sensitivity of whatever's inboard of the cell
is unknown a priori.
In custom analog land, you might as well forget
all this. Even for a structured ASIC or standard
cell library, for this approach to work requires
additional design style constraint and/or per-
pin characterization to prove the validity of ignoring "should be same"
As we unveil EE Times’ 2015 Silicon 60 list, journalist & Silicon 60 researcher Peter Clarke hosts a conversation on startups in the electronics industry. Panelists Dan Armbrust (investment firm Silicon Catalyst), Andrew Kau (venture capital firm Walden International), and Stan Boland (successful serial entrepreneur, former CEO of Neul, Icera) join in the live debate.