• Previous message: R. Timothy Edwards: "Extraction of non-Manhattan geometry"
• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

All:

I would like to add another perspective to the question of a graphical
user interface to magic.

First, in my opinion it is the underlying data structure (corner
stitched data structure) and the well-thought out graphics which make
magic superior to the closed source layout editors. I consider it a key
insight by Osterhout et al. that in chip layout the artificial breakup
of a mask into "polygons", "paths", "rectangles", etc. just add noise to
the displayed picture which is irrelevant to the functioning of the
final circuit. "MacPaint" results in a better layout tools than
"MacDraw". I also greatly appreciate Tim Edward's effort at
non-Manhattan geometry. In sub-0.25um layout there are not too many of
these shapes (mostly in the pad ring), but not having non-Manhattan
geometry has excluded magic from being considered for serious
applications.

Second, OpenGL on Linux is by now very well supported within XFree86,
for free (for a few graphics chips at least). Alternatively, even for a
cheap NVIDIA TNT2 based video card there are very good drivers available
from NVIDIA (closed source Linux) for free. I have them running on
several PC's (from my 6-year old's 300MHz PIII to dual-processor 1GB
workstation) with no problems at all (except a detail in magic, see my
previous post).

Third, there are two factors which work in favor of adding complexity to
magic for the benefit of useability at the expense of more computing
power required - Moore's law means that every 2 years or so the price of
DRAM reduces by 50%, and the CPU speed increases 2x; and Microsoft
software which every 2 years requires double the RAM and the CPU speed
and raises the bar on average PC performance. I work in R&D at a major
semiconductor supplier and can attest that Moore's law will hold past
the 0.1um generation. So, I really liked Jeff Solomon's paper at DAC
(http://www.dac.com/39th/talkindex.html, paper 31.2, need Windows
Mediaplayer) about trading RAM for display speed. This was a very
elegant paper.

Fourth, the GUI is not Microsoft's invention. I consider KDE to be the
best GUI available for any computer today, mainly because they try to
make the UNIX desktop more useable. So, the pinnacle of achievement
would be to port magic to KDE (my opinion, other people may prefer
GNOME). I am working myself into the internals but progress is slow...
Imagine having the first fully internationalized layout editor, e.g.
Chinese menues, opening up chip layout to another 1B potential users.

Fifth, I prefer python over Scheme since most casual computer users or
even engineers can deal with a procedural language (like Perl or Python)
but are lost with lisp-like languages. Python seems to be the language I
can use every two months without having to re-learn unnatural syntax.
I'll be the first language I'll teach my 6-year old (who has used magic
once or twice to "help" me with test chip layout).

So, I think the ultimate layout tool would be magic in KDE with Python
as scripting language. I'll start working on it over Christmas...

Best regards,

Erwin

• Next message: Spaceborne Calin: "Re: reply to Calin on "magic-wrapping""
• Previous message: R. Timothy Edwards: "Extraction of non-Manhattan geometry"
• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]