ALF meeting, July 19 @ Synopsys, Mountain View

Attendees:

Dhaval Sejpal		IBM
Nicholas Barbieri	Synopsys (morning)
Fereshte Kianersi	Synopsys
Shir-Shen Chang		Synopsys
Tim Jennings		Philips
Tim Ehrler		Philips
John Peters		Philips
Greg Dufour		Mentor
Carmel Collins		Artisan (afternoon)
Rajesh Uppuluri		Logicvision (by phone)
Dan Moritz		LSI Logic (afternoon)
Wolfgang Roethig	NEC

The following chapters of the ALF 2.0 spec were reviewed.
Section 10: "Physical Modeling"
Section 5:  "Functional Modeling"
Section 6:  "Modeling for Test"
Section 7:  "General Rules for Arithmetic Models"
Section 8:  "Electrical Performance Modeling" (preliminary)

Section 10.2
Arithmetic models in the context of layout
------------------------------------------
The legal range of values (number, unsigned, non-negative number etc.)
is specified for each model.
Shir-Shen requested to change non-negative numbers to positive numbers
for certain models

A.I. propose models for which the change from non-negative numbers
     to positive numbers applies - Shir Shen
DONE - see separate note

Section 10.3
Statements for geometric transformations
----------------------------------------
Need to change "mathematical positive sense" into "counter-clockwise"

Shir-Shen pointed out the necessary range of values for ROTATE and FLIP
 0 < FLIP   <  180
 0 < ROTATE <  360

A.I. Incorporate changes into spec

Section 10.5
LAYER statement
---------------
PURPOSE annotation:
A.I. introduce annotation value "reserved" for layer with proprietary use only

Section 10.6
Geometric model statement
-------------------------
POINT_TO_POINT annotation
A.I. provide illustration for annotation value "rectilinear"

TEMPLATES for predefined geometric models:
A.I. Make clear that RECTANGLE and LINE are predefined,
     HORIZONTAL_LINE etc. are user-defineable

A.I.Introduce VERTEX annotation for EXTENSION
	VERTEX = round | straight
	default is straight
provide illustration

Section 10.8
VIA statement
-------------
Proposal: stacked vias may contain more than 3 patterns,
restriction to 3 pattern (lower metal, cut, upper metal)
is for backward compatibility with LEF, but not in the language

USAGE Annotatin value "stack_only" is for backward compatibility with LEF
Do we need new USAGE value for completely described stack?
"partial_stack" instead of "stack_only"?
"full_stack" to describe the complete stack (more than 3 patterns)?

need example for via reference

Section 10.9
BLOCKAGE statement
------------------
Issue: How can soft blockages be described?
This is an hierarchical design issue.

Section 10.11
RULE statement
--------------
Distance is measured with respect to non-extended object,
i.e., center of a line etc.
Do we need SPACING to distinguish edge-to-edge measurements
from center-to-center measurements or will an annotation for
DISTANCE suffice?
Proposal: introduce REFERENCE annotation for DISTANCE
	REFERENCE = center | edge
	which one should be default?
Alternative proposal:
	DISTANCE = center to center
	SPACING  = edge to edge

Section 6.1
Annotations and attributes for a CELL
-------------------------------------
Proposal:
Add ATTRIBUTE "GPU" = Graphics Processor Unit
for cell with CELLTYPE=core

A.I. Add diagram for swap_class, restrict_class
to illustrate the example

Note on BUFFERTYPE:
not only for cells with CELLTYPE=buffer, but also for
complex I/O cells, for instance I/O flipflop
BUFFERTYPE annotation is prerequisite for DRIVERTYPE annotation

Section 6.2
NON_SCAN_CELL statement
-----------------------
A.I. put old NON_SCAN_CELL syntax into phased-out items

How to decide whether a cell has non-scan-cell:
Rule: A cell with restrict_class { scan } only
must be a scan cell with a defined non_scan_cell replacement

Question: Should we introduce SCAN_CELL statement also
to allow a two-way x-reference between scan and non-scan cell?
A.I. State opinion - DFT tool vendors (Synopsys, Mentor, LogicVision)

Section 6.4
Annotations and attributes for a PIN
------------------------------------
Clarification on SCAN_POSITION
A.I. change meaning of annotation value "0" to be compatible with OLA.
Value "0" shall mean, pin is not on scan chain.

Section 6.6
Annotations for other objects
-----------------------------
OPERATION annotation value "iddq" applies for cell with any celltype,
as opposed to memory-specific operation values.

Section 7.4
Annotations for arithmetic models
---------------------------------
A.I. specify that the value of UNIT should be a positive number

A.I. specify extrapolation behavior for INTERPOLATION annotation
Proposal:
	Extrapolation with "fit": use two outer datapoints
	Extrapolation with "floor" or "ceiling" out of range:
	use smallest or largest datapoint, respectively

All A.I. for spec edits are for Wolfgang and Joe Daniels