Subject: Re: Port vs. Process bound semantics.
From: Kevin Cameron (dkc@galaxy.nsc.com)
Date: Fri Dec 21 2001 - 14:39:21 PST
Just to recap. My view of the Port vs. Process approaches is that from a modelling
accuracy perspective process-bound should apply unless a port has a disrupting
effect to the electrical characteristics of the net. I.e. if the net is one physical
wire (e.g. a single piece of metal) from all drivers to receivers then the user
should be specifying "process bound". If the port breaks its top and bottom sides
then the top side and bottom side nodes need to be handled seperately in simulation;
this would be the case if the digital design signal crosses (say) a pad boundary
which can be considered transparent logically but in reality has seperate connections
to the external and internal modules.
If a port-bound semantic is being used but there is only one port boundary between
any driver and receiver and the top of the net hierarchy (e.g. all processes are in leaf
cells) it is a degenerate case that will be equivalent to using the process-bound
semantics. The difference appears when there is mixed behavior on both sides of the
port, in which case the top and bottom nets need to resolve their disciplines and signal
values seperately. The latter case would occur when (say) a "black box" digital
module is used in an analog chip in a digitally described board-level talking to another
similar chip: the digital nets in the "black box" and at the board level are "digital
islands" (in Jon's parlance). Similarly a digital chip with say an analog PLL connecting
through a layer of digital to an analog described board level would give rise to two
seperate analog nodes (one in the PLL and one for the PCB track).
Discipline resolution for attributes, merging and splitting apply to the drivers and
receivers of each "wire segment" seperately, i.e. in the "black box" case above
you would not merge drivers from the board level with the "black box" level.
When process-bound and port-bound are mixed the process bound drivers and
receivers are considered attached to the wire segment of their instance's ports,
i.e. port-bound net partitioning takes precedence.
Regards,
Kev.
This archive was generated by hypermail 2b28 : Fri Dec 21 2001 - 14:32:55 PST