( ESNUG 428 Item 7 ) -------------------------------------------- [04/28/04]
Subject: ( ESNUG 427 #9 ) Nassda's Follow-up To The Lexsim User Review
> Anonymous, please. We use Lexsim to analyze the reliability of our power
> network. It takes in a netlist from both the Synopsys and Mentor RC
> extraction tools. Our power network is huge. Lexsim uses a special
> algorithm to simplify it into smaller network -- basically it is an
> impedance-based network. Then it calculates how much current is drawn,
> and brings the current information into the network & calculates IR drop.
>
> - [ The Mouse That Roared ]
From: Simon Young <spike=user domain=nassda spot calm>
Hi, John,
By running Lexsim in the coupled mode (that is, with reduced power net
back-annotated), 'Mouse' has also gained one piece of information vital
to nanometer design; the timing impact of the dynamic power net variation.
As process geometries are shrinking, threshold voltages don't scale as
quickly as VDD. For example, start with a 1V VDD, take off two Vt (the
best-case situation, for an inverter P/N stack), and there is little
headroom left. This sensitivity significantly increases timing variation.
I've seen 40% access time deviation for some large static RAMs, induced
by dynamic voltage drop.
Calculating dynamic voltage drop and current density helps to determine
reliability -- but friends don't let friends ignore the timing impact.
> There are two things a Lexsim user needs to pay special attention to:
>
> 1) The vector generation for the Lexsim. The vectors should cover
> worst case scenario of toggling the maximum number of nodes.
Agreed. After almost a decade working in voltage drop analysis, I can say
that this is one of the unsolved challenges. The QoR for any voltage drop
analysis is almost entirely determined by the precision of the currents
used (consider the errors and their magnitude in V' = V - IR.) Lexsim is
targeted at memories and analog/mixed-signal circuits, as the constraints
on the circuit and the creation of vectors by the designer is manageable.
Doing coupled analysis at the transistor-level for synthesized logic,
(which is needed for accurate currents and good QoR) is very difficult
and requires new technology.
> My biggest criticism is with regards to Lexsim RA. With Lexsim alone,
> the IR drop information is presented in a text file. If you want to
> export the information into a viewer (like Cadence), you must also buy
> from Nassda a copy of Lexsim RA.
Because eveyone else in EDA is doing platforms, on April 5th Nassda
announced the HSIMplus platform, which includes the Lexsim technology
packaged as new HSIMplus options. HSIMplus v5.0s ischeduled to be released
by May, and the RA capability is now included with the IR drop analysis.
- Simon Young
Nassda Corp. Santa Clara, CA
|
|
