[ibis-interconn] Re: What do we want an IBIS (.ibs) file to represent?

  • From: "Lynne C. Green" <lgreen22@xxxxxxxxxxxxxx>
  • To: IBIS-Interconnect <ibis-interconn@xxxxxxxxxxxxx>
  • Date: Mon, 22 Oct 2012 15:20:22 -0700

All,

One small thought:

* Since different chips can go in a package, "signal name" could change with each chip. Assigned pins for power/gnd/*ref could also change with each chip. If this is to be a package model independent of the chip/PCB, could "signal name" be dropped? Could this break EDA tools/flow?

* An interesting extension: the concept of a check for "continuity" of power/gnd/*ref/ between component and die, which might be useful for the end user. If die pad (chip design) and signal name (PCB) always match, could these also be continuity checked?

* Note that "stand-alone" package models sometimes do not include the bond wire. The bond wire length (and thus parasitics) sometimes depends on the chip size. Is there some way to bridge this gap, either through "comment" documentation or bond wire subparam(s)?

Cheers,
Lynne


On 10/22/2012 11:03 AM, Walter Katz wrote:

All,

I think we can all agree that an Integrated Circuit "Part" consists of a die inside of a package (lets puts MCM, DIMM aside for now).

I think we can also agree that it is common for a die to be used in multiple packages, and a package can be used for multiple dies. IBIS supports the former (a single .ibs file can have multiple [Components]), but not the later. The pin_name, signal_name, model_name (A1 DQ1 DQ) discussion during last week's meeting was at the heart of this matter.

I think we should agree that evolving IBIS so that it treats the die and package "independently" is the direction we should be going. Die centric is signal_name, model_name (DQ1 DQ) while package centric is pin_name, signal_name(A1 DQ1).

We can do this maintaining IBIS backward compatibility and minimal impact on IC Vendors, EDA Vendors and Users! We do this by making the following statements about IBIS:

1.[Pin]

a.Contains a list of all of the connects of the [Component] to the "Board"

b.There are three categories of [Pin] model_name

i.NC

ii.POWER and GND

1.Each unique signal_name has

a.one implied die pad name

b.may have a list of named die pad names

iii.Signal (not NC, POWER or GND)

1.Each unique signal_name corresponds to one of:

a.Unique instance of a single ended Input, Output, or I/O Buffer model

b.One side of a differential Input, Output, or I/O Buffer model

c.Unique instance of a Terminator Model

2.[Die Pad]

a.New Optional Keyword

b.Implicit Die Pad pad_names are

i.POWER and GND unique signal_names

ii.Each unique signal_name when [Pin] model_name is not NC, POWER or GND

c.Named Die Pad pad_names are

i.For signal_names on [Pin]s with model_name POWER or GND

3.[Model] supply keywords

a.PuRef <signal_name>

i.Pullup Reference signal_name

b.PdRef <signal_name>

i.Pulldown Reference signal_name

c.PcRef <signal_name>

i.Power Clamp Reference signal_name

d.GcRef <signal_name>

i.GND Clamp Reference signal_name

Package Interconnect Model ports are [Pin] pin_names and [Die Pad] pad_names.

On-Die Interconnect Model ports are [Die Pad] pad_names, [Model] signal ports, and [Model] supply ports.

Just as IBIS currently has a method of associating package models with both specific pins and a default for all pins, the new package and on-die modeling should have a method of not only assigning ports of IBIS-ISS subckts to specific [Pin] pin_names, [Die Pad] pad_names, [Model] signal ports, and [Model] supply ports, there should also be methods to assign default package and on-die models to [Model]s.

Walter

Walter Katz

wkatz@xxxxxxxxxx <mailto:wkatz@xxxxxxxxxx>

Phone 303.449-2308

Mobile 303.335-6156


Other related posts: