pi: prof. jason cong (ucla) graduate students: chin-chih chang, david pan, xin yuan

Interconnect Planning, Synthesis, Interconnect Planning, Synthesis, and Layout for Performance, Signal and Layout for Performance, Signal

Reliability and Cost OptimizationReliability and Cost Optimization

SRC Task ID: 605.001SRC Task ID: 605.001

PI: Prof. Jason Cong (UCLA)Graduate Students: Chin-Chih Chang, David Pan, Xin Yuan

Industrial Liaisons: Dr. Prakash Arunachalam (Intel) Dr. Norman Chang (HP)

Dr. Wilm Donath (IBM) Dr. Stefan Rusu (Intel)SRC Monitor: Lawrence Arledge (SRC)

Project Overview

Objective: investigate an interconnect-centric

design flow and methodology, consisting of:

Interconnect Planning

Interconnect Synthesis

Interconnect Layout

Overview of Interconnect-Centric IC Design Flow

Architecture/Conceptual-level Design

Design Specification

Final Layout

abstractionStructure viewFunctional viewPhysical viewTiming view

HDM

Synthesis and Placement under Physical Hierarchy

Interconnect Planning•Physical Hierarchy Generation•Foorplan/Coarse Placement with Interconnect Planning•Interconnect Architecture Planning

Interconnect Optimization (TRIO)

• Topology Optimization with Buffer Insertion• Wire sizing and spacing• Simultaneous Buffer Insertion and Wire Sizing• Simultaneous Topology Construction with Buffer Insertion and Wire Sizing

Interconnect LayoutRoute Planning

Point-to-Point Gridless Routing

Interconnect Performance Estimation Models (IPEM)

•OWS, SDWS, BISWS

Interconnect SynthesisPerformance-driven Global Routing

Pseudo Pin Assignment under Noise Control

Overview of Interconnect-Centric IC Design Flow

Architecture/Conceptual-level Design

Design Specification

Final Layout

abstractionStructure viewFunctional viewPhysical viewTiming view

HDM

Synthesis and Placement under Physical Hierarchy

Interconnect Planning•Physical Hierarchy Generation•Foorplan/Coarse Placement with Interconnect Planning•Interconnect Architecture Planning

Interconnect Optimization (TRIO)

• Topology Optimization with Buffer Insertion• Wire sizing and spacing• Simultaneous Buffer Insertion and Wire Sizing• Simultaneous Topology Construction with Buffer Insertion and Wire Sizing

Interconnect LayoutRoute Planning

Point-to-Point Gridless Routing

Interconnect Performance Estimation Models (IPEM)

•OWS, SDWS, BISWS

Interconnect SynthesisPerformance-driven Global Routing

Pseudo Pin Assignment under Noise Control

Interconnect Synthesis

Performance-driven Global Routing

Pseudo Pin Assignment under Noise Control

Interconnect LayoutRoute Planning

Point-to-Point Gridless Routing

Interconnect Performance Estimation Models (IPEM)

• OWS• SDWS• BISWS

Interconnect Optimization (TRIO)

• Topology Optimization with Buffer Insertion• Wire sizing and spacing• Simultaneous Buffer Insertion and Wire Sizing• Simultaneous Topology Construction with Buffer Insertion and Wire Sizing

Interconnect Planning

• Physical Hierarchy Generation• Foorplan/Coarse Placement with Interconnect Planning• Interconnect Architecture Planning

Overview of Interconnect-Centric IC Design Flow

Architecture/Conceptual-level Design

Design Specification

Final Layout

abstractionStructure viewFunctional viewPhysical viewTiming view

HDM

Synthesis and Placement under Physical Hierarchy

Interconnect Planning•Physical Hierarchy Generation•Foorplan/Coarse Placement with Interconnect Planning•Interconnect Architecture Planning

Interconnect Optimization (TRIO)

• Topology Optimization with Buffer Insertion• Wire sizing and spacing• Simultaneous Buffer Insertion and Wire Sizing• Simultaneous Topology Construction with Buffer Insertion and Wire Sizing

Interconnect LayoutRoute Planning

Point-to-Point Gridless Routing

Interconnect Performance Estimation Models (IPEM)

•OWS, SDWS, BISWS

Interconnect SynthesisPerformance-driven Global Routing

Pseudo Pin Assignment under Noise Control

Review: Accomplishments in Year 1& 2

Efficient (constant time) and accurate (90%) interconnect delay estimation models for 2-pin nets under different interconnect optimization algorithms [Cong-Pan, IWLS’98, SRC/TECHCON’98,

ASPDAC’99]

Interconnect architecture planning [Cong-Pan,DAC’99]

Efficient and accurate interconnect estimation models for multiple-pin nets [Cong-Pan, TAU’99]

Buffer block planning for interconnect-driven floorplanning [Cong-Kong-Pan, ICCAD’99]

Accomplishments and Ongoing Works in Year 3

An improved crosstalk model with application to noise constrained interconnect optimization. [Cong-Pan-Srinivas, SRC Techcon’00, TAU’00]

Pseudo pin assignment with crosstalk noise control [Chang-Cong, ISPD’00]

Routing tree construction under fixed buffer locations [Cong-Yuan, DAC’00]

Ongoing studies on physical planning

Accomplishments and Ongoing Works in Year 3

An improved crosstalk model with application to noise constrained interconnect optimization. [Cong-Pan-Srinivas, SRC Techcon’00, TAU’00]

Pseudo pin assignment with crosstalk noise control [Chang-Cong, ISPD’00]

Routing tree construction under fixed buffer locations [Cong-Yuan, DAC’00]

Ongoing studies on physical planning

Crosstalk Noise

0

0.2

0.4

0.6

0 0.5 1 1.5 2 2.5

Time (ns)

Vo

ltag

e S

pik

e (V

)

Cx

Victim net

Aggressor net

Previous Works Transmission line equations [Sakurai+, TED’93, ASPDAC’98]

Only handle fully coupled bus lines Devgan’s model [ICCAD’97]

Elegant, Elmore-like formula for peak noise Over estimation, esp. when aggressor slew is small => could lead

to noise even larger than Vdd ! Charge-sharing based (e.g., [Vittal & Marek-Sodawska,

TCAD’97]) One lumped R, C for victim/aggressor net Simple noise formulae for peak noise, and noise amplitude-width

product Cannot differ near-source versus near-sink coupling

Need simple yet accurate model that considers more KEY (but not more than necessary) parameters to guide layout optimization !

2- Crosstalk Noise Model [Cong-Pan-Srinivas, SRC Techcon’00,

TAU’00]

Cx

Victim net

Aggressor net

Ls Le Cl

Cs1

Tr

Rd Rs

Ce1Cs2

Ce2Cl

Re

Lc

2- Model

C1=Cs1

Rd Rs

C2=Cs2+Ce1 CL=Ce2+Cl

Re

Cs1

TrRd Rs

Ce1Cs2

Ce2Cl

Re

Closed-Form Solutions

Peak noise

vr

r

x ttet

tv /1max

vr

vr

vrwidthtte

ttettt

/1

/21ln

Let1e12 R))((

)(

CCRCCCxRRt

CRRt

dLsdv

xsdx

tx: RC delay from upstream resistance times coupling cap.tv: Elmore delay of the victim net

Noise widthNoise width

Unified View for Existing Models

Peak noise

vr

r

x ttet

tv /1max

As v

xr

t

tvt max , then0

(Vittal+ TCAD’97 model)

As r

xvr

t

tvtt max , then (Devgan ICCAD’97 model)

(2- model)

221

1

1

2

11/1max

rv

x

v

rv

x

v

r

v

xvr

r

x

tt

t

ttt

t

t

t

t

tttet

tv

As

(Vittal+ TCAD’99 model,Up to 100% larger than Devgan metric for large tr)

Experimental Results

1,000 random nets based on realistic parameters

Average Error (%)

0

100

200

300

400

500

600

700

devgan vittal 2-Pi

Model

Perc

enta

ge

Average Error (%)

0

2

4

6

8

10

12

14

16

18

vittal 2-Pi

Model

Perc

enta

ge

Applications of 2- Model

We have obtained a set of rules for noise reduction using different interconnect optimizations Driver sizing Near source versus sink coupling Shield insertion Wire sizing and spacing AW product (noise amplitude • width)

Used in Magma’s BlastFusion software -- U.S. Patent Pending

Accomplishments and Ongoing Works in Year 3

An improved crosstalk model with application to noise constrained interconnect optimization. [Cong-Pan-Srinivas, SRC Techcon’00, TAU’00]

Pseudo pin assignment with crosstalk noise control [Chang-Cong, ISPD’00]

Routing tree construction under fixed buffer locations [Cong-Yuan, DAC’00]

Ongoing studies on physical planning

Overview of Interconnect-Centric IC Design Flow

Architecture/Conceptual-level Design

Design Specification

Final Layout

abstractionStructure viewFunctional viewPhysical viewTiming view

HDM

Synthesis and Placement under Physical Hierarchy

Interconnect Planning•Physical Hierarchy Generation•Foorplan/Coarse Placement with Interconnect Planning•Interconnect Architecture Planning

Interconnect Optimization (TRIO)

• Topology Optimization with Buffer Insertion• Wire sizing and spacing• Simultaneous Buffer Insertion and Wire Sizing• Simultaneous Topology Construction with Buffer Insertion and Wire Sizing

Interconnect LayoutRoute Planning

Point-to-Point Gridless Routing

Interconnect Performance Estimation Models (IPEM)

•OWS, SDWS, BISWS

Interconnect SynthesisPerformance-driven Global Routing

Pseudo Pin Assignment under Noise Control

Coupling: 4 Coupling: 2

Pseudo Pin Assignment with Crosstalk Noise Control

Pseudo pin: a point where a net crosses a tile boundary Pseudo pin assignment: bridge between global routing

and detailed routing Our contributions: Pseudo pin assignment algorithm

for gridless general area routing with noise control Control crosstalk noise Handle obstacle constraints Align pseudo pins for detailed routing routability Reduce the total wire length

Vias: 6 Vias: 8

Why Crosstalk Noise Control in Pseudo Pin Assignment

What can we do in routing to affect crosstalk? Buffer insertion (if the global router does it) Wire ordering Wire spacing

Determine wire ordering and spacing Global routing? High complexity, hard to consider obstacles Detailed routing? Flexibility is low Pseudo pin assignment? Reasonable complexity and high

accuracy

PPA Algorithm Overview

One layer at a time Optimize one row at a time

Maximum strip boundary decomposition:

Partition boundary to intervals

Coarse pseudo pin assignment:

Assign pseudo pins to intervals

Detailed assignment within each “strip”:

Determine pseudo pin locations

A Noise Distribution Example – After Detailed Routing

0500

10001500200025003000350040004500

nets

0-0.1 0.1-0.2

0.2-0.3

0.3-0.4

0.4-0.5

noise (Vdd)

no noise control

with noisecontrol

Test case: scaled mcc2, NTRS’97 0.18 um Tech, 0.3 Vdd noise constraints

Pseudo pin assignment with noise control effectively reduce crosstalk noise and meet noise constraints

Accomplishments and Ongoing Works in Year 3

An improved crosstalk model with application to noise constrained interconnect optimization. [Cong-Pan-Srinivas, SRC Techcon’00, TAU’00]

Pseudo pin assignment with crosstalk noise control [Chang-Cong, ISPD’00]

Routing tree construction under fixed buffer locations [Cong-Yuan, DAC’00]

Ongoing studies on physical planning

Motivations for Routing Tree Construction under Fixed Buffer Locations

Given buffer blocks planned in early stage, how to do routing and buffer insertion under fixed buffer locations?

Investigate different buffer block planning schemes.

An Example of Floorplan with Buffer Block Planning

sink

hard block or IP

buffer blockobstacle

source

RMP (Recursively Merging and Pruning) Algorithm

Basic Idea: A bottom-up tree construction combined with

buffer insertion. Generate a set of subtrees from sinks and then

gradually expand and merge them until a complete tree with best performance is produced.

Key differences from previous approaches The sets of subtrees may not be disjoint Multiple subtrees may be generated at a node Works on a routing graph Handle multiple-pin nets with fixed buffer

locations constraints.

Comparison BetweenRMP and Modified BA-tree Algorithm

Modify BA-tree algorithm [TRIO] to handle fixed buffer insertion (MBA-tree) Similar to the semi-automatic approach used in real design: “round” ideal buffers in

BA-tree to the given buffers in MBA-tree.

#pinsRMP MBA-tree

(adjacent NR)MBA-tree

(t-adjacent NR)rat wl rat wl rat wl

4 1.0 1.0 1.62 0.90 1.29 1.08

5 1.0 1.0 1.99 0.90 1.46 1.13

6 1.0 1.0 1.84 0.85 1.29 0.97

Experimental results of RMP vs. BMA-tree (all data are normalized with respect to RMP).

RMP can outperform MBA-tree by up to 50% in terms of delay with comparable wirelength.

Accomplishments and Ongoing Works in Year 3

An improved crosstalk model with application to noise constrained interconnect optimization. [Cong-Pan-Srinivas, SRC Techcon’00, TAU’00]

Pseudo pin assignment with crosstalk noise control [Chang-Cong, ISPD’00]

Routing tree construction under fixed buffer locations [Cong-Yuan, DAC’00]

Further study on physical planning

Ongoing Studies on Physical Planning

Motivation Logical hierarchy is different to physical hierarchy Planning based on logical hierarchy is limited by

floorplanning on logical hierarchy Planning based on physical hierarchy with

performance driven geometric embedded partitioning shows good promises [Cong-Lim ICCAD’00]

Further study on interconnect planning in physical hierarchy

Example of Logic Hierarchy in Final Layout

By courtesy of IBM (Tony Drumm)

Example of Logic Hierarchy in Final Layout

By courtesy of IBM (Tony Drumm)

Several Ongoing Efforts

Studies on the impacts of layer assignment

Studies on the congestion of global interconnect

A reasonable formulation for physical planning

Impacts of Layer Assignments on Delays

0

200

400

600

800

1000

1200

1400

1600

wire length (mm)

dela

y (p

s)

layer 1-2layer 3-4layer 5-6layer 7-8

Delays estimated by IPEM for optimal buffer insertion and wire sizing

0.13 um NTRS’97 technology Delay reduction of changing wires

from layers 1-2 to layers 7-8: 0.5mm: 2% 6.5mm: 40% 27.5mm: 49%

The longer the wire, the more possible reduction by assigning to upper metals

Several Ongoing Efforts

Studies on the impacts of layer assignment

Studies on the congestion of global interconnect

A reasonable formulation for physical planning

Congestion Analysis – Efforts and Preliminary Results

Upper metal layers are used for long wires for reducing delays

Are there enough routing resource on upper metal layers? What is the trend? How many long wires that need to be routed on upper metal

layers? How much space available on upper metal layers (Power and

Clock nets are also competing)? List of industrial contacts that we consulted:

IBM – Tony Drumm, John Darringer Intel – Desmond Kirkpatrick, Mosur Mohan, Kris Konigsfeld HP – Norman Chang

Layer Competition under Different Target Delays (Test case from IBM)

Wire area requirements under different target delays using IPEM with 0.13um NTRS'97 Tech

050

100150200250300350

target delay(ps)

area

delay not meetlayer 7-8 or abovelayer 5-6 or abovelayer 3-4 or abovelayer 1-2 or above

Several Ongoing Efforts

Studies on the impacts of layer assignment

Studies on the congestion of global interconnect

A reasonable formulation for physical planning

Problem Formulation for Physical Planning

Plan the following to achieve optimization goals: Physical hierarchy with rough geometric information for

identifying global interconnects Retiming and pipelining Layer assignment for global interconnects Global net topology generation and congestion control Buffer planning …

Estimate and Optimize the following objectives: Delay Area Power Signal integrity

Deliverables Development of efficient and accurate interconnect performance

estimation models for interconnect-driven synthesis and planning (Completed - 30-Jun-1999)

Development of interconnect architecture planning framework (Completed - 30-Jun-1999)

Development of efficient algorithms for integrated interconnect planning & floorplanning capabilities at the physical level (Completed - 30-Sep-1999)

Development & validation of accurate noise models to guide the interconnect synthesis algorithm for signal reliability (Completed - 31-Dec-1999)

Development of optimal or near-optimal interconnect synthesis algorithm for multiple spatially or temporally related signal nets for performance & signal reliability optimization (Completed - 31-Dec-1999)

Development of efficient algorithms for integrated interconnect planning & floorplanning capabilities at the RTL-level; Software (Planned - 31-Dec-2000)

Technology Transfer TRIO (Tree-Repeater-Interconnect-Optimization) package

Integrated into Intel design technology http://cadlab.cs.ucla.edu/~trio

IPEM (Interconnect Performance Estimation Model) package Integrated into IBM design technology http://cadlab.cs.ucla.edu/software_release/ipem/htdocs

Wire width planning U.S. Patent pending under SRC sponsorship

BBP (Buffer Block Planning) for physical level floorplanning Source code transferred to IBM Interests from Intel and HP

Crosstalk noise modeling U.S. Patent pending (joint with Magma)

Summary

An improved crosstalk model with application to noise constrained interconnect optimization.

Pseudo pin assignment with crosstalk noise control

Routing tree construction under fixed buffer locations

Ongoing studies on physical planning

Milestones Development of a computational model for interconnect architecture planning based

on a given design characterization (specified in terms of target clock rate, interconnect distribution, depths of logic,network, etc.) (31-Dec-1998)

Development of estimation models for interconnect layout optimizations suitable for pre-layout synthesis and planning (31-Dec-1998)

Development of efficient algorithms for integrated interconnect planning and floorplanning capabilities at the RTL-level (31-Dec-1999)

Completion of the ongoing effort on the development on a multi-layer general-area gridless routing system (31-Dec-1999)

Development of optimal or near-optimal interconnect synthesis algorithm for multiple spatially or temporally related signal nets for performance and signal reliability optimization (31-Dec-1999)

Development and validation of very efficient but accurate noise models to relate the noise with the physical parameters to guide the interconnect synthesis algorithm for signal reliability optimization (31-Dec-1999)

Development of efficient algorithms for integrated interconnect planning and floorplanning capabilities at the physical level (31-Dec-1999)

Development of efficient algorithms for integrated interconnect planning and floorplanning capabilities at the RT-level (31-Dec-2000)