9/11/2015 1 VLSI Physical Design Automation Prof. David Pan [email protected] Office: ACES 5.434...

33
03/27/22 1 VLSI Physical Design Automation Prof. David Pan [email protected] Office: ACES 5.434 Misc. Topics and Conclusion

Transcript of 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan [email protected] Office: ACES 5.434...

Page 1: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

04/19/23 1

VLSI Physical Design Automation

Prof. David Pan

[email protected]

Office: ACES 5.434

Misc. Topics and Conclusion

Page 2: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

204/19/23

Other Design Styles: FPGA• Field Programmable Gate Array• First introduced by Xilinx in 1984.• Pre-fabricated devices and interconnect, which are

programmable by user.• Advantages:

– short turnaround time.– low manufacturing cost.– fully testable.– re-programmable.

• Particularly suitable for prototyping, low or medium-volume production, device controllers, etc.

Page 3: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

304/19/23

Comparison of Design Styles

Full-CustomStandard

CellGate Array FPGA

Cell size variable fixed height fixed fixed

Cell type variable variable fixedprogramma

ble

Cell placement variable in row fixed fixed

Interconnections variable variable variableprogramma

ble

Fabrication layers

all layersall

layersrouting

layers only no layers

Page 4: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

404/19/23

Comparison of Design Styles

Full-Custom Standard Cell Gate Array FPGA

Area compactcompact to moderate

moderate large

Performance highhigh

to moderatemoderate low

Design cost high medium medium low

Time-to-market long medium medium short

Page 5: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

504/19/23

Programming Technologies

• SRAM to control pass transistor / multiplexer• EPROM – UV light Erasable PROM• EEPROM – Electrically Erasable PROM• Antifuses – One time programmable

• They are different in ease of manufacturing, manufacturing reliability, area, ON and OFF resistance, parasitic capacitance, power consumption, re-programmability.

Page 6: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

604/19/23

Typical FPGA Architecture• Consists of: Logic modules, Routing resources, and

I/O modules.

Logic ModuleLogic Module

IO ModuleIO Module

Routing TracksRouting Tracks& Switch boxes& Switch boxes

Page 7: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

704/19/23

FPGA Architecture Examples

LogicLogicModuleModule

Array-based ModelArray-based Model Row-based ModelRow-based Model

Sea-of-Gates ModelSea-of-Gates Model Hierarchical ModelHierarchical Model

RoutingRoutingresourcesresourcesoverlayedoverlayedon logicon logicmodulesmodules

Page 8: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

804/19/23

Two Types of Logic Modules

Look-Up Table (LUT) based:• A block of RAM to store the truth table.• A k-input 1-output functions needs 2k bits.• k is usually 5 or 6.

Multiplexer based: e.g., f=ABC+ABCe.g., f=ABC+ABC

CCBBAA

AABB

ff

Page 9: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

904/19/23

Two Types of Switchboxes• First Type:First Type:

• Second Type:Second Type:

Page 10: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

1004/19/23

Several Segmentation Models• Non-Segmentation Model:

• Uniform Segmentation Model:

11 44 00 00 22 00 00 33 00 55 00 00 00 00

00 00 00 11 00 00 44 00 22 00 33 00 00 55

ConnectingConnecting

Not connectingNot connecting

11 44 00 00 22 00 00 33 00 55 00 00 00 00

00 00 00 11 00 00 44 00 22 00 33 00 00 55

Fuse orFuse orProgrammableProgrammable

switchswitch

Page 11: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

1104/19/23

Several Segmentation Models

• Uniform Staggered Segmentation Model:

• Non-uniform Staggered Segmentation Model:

11 44 00 00 22 00 00 33 00 55 00 00 00 00

00 00 00 11 00 00 44 00 22 00 33 00 00 55

11 44 00 00 22 00 00 33 00 55 00 00 00 00

00 00 00 11 00 00 44 00 22 00 33 00 00 55

Page 12: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

1204/19/23

Comparison of Segmentation Models

• The segmented model provides better utilization of routing resources.

• However, segmented model uses more fuses or programmable switches.

• The delay of a net is directly proportional to the # of fuses or programmable switches in the route– Manhattan-distance based delay model does NOT work

anymore– The segmented model is slower in general

Page 13: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

1304/19/23

Physical Design of FPGAs

• Very different from other design styles• Architecture dependent:

– LUT or Multiplexer in logic modules– Type of switchboxes used– Type of segmentation model used– ......

• Physical Design:– Partitioning– Floorplanning/Placement– Routing

Page 14: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

1404/19/23

Partitioning

• Want to partition the circuit such that each partition (cluster) can be implemented by a logic module.

• Also called Clustering.• # of I/O pins, not cluster sizes, is important. (For multiplexer based logic modules, functionality of

clusters is also important.)

Example:Example:Using 4-LUTsUsing 4-LUTs

Page 15: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

1504/19/23

Placement

• Assign clusters formed during partitioning to logic modules of FPGA.

• The problem is the same as gate-array placement.

Page 16: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

1604/19/23

Routing

• Global routing:– Similar to global routing in other design styles.– Minimize wire length and balance densities.

• Detailed routing:– Very different from other design styles.– Different algorithms for different segmentation models.– Channels and switchboxes have fixed capacities.

Page 17: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

1704/19/23

Structured ASIC

• New buzz word, but essentially gate array– Mask reconfigurable – Not field reconfigurable

• Between FPGA and standard cells– Balance delay/performance and mask cost

• Only programmable once– by vias (e.g., Via-Programmable Gate Array – VPGA)

Page 18: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

04/19/23 18

Physcial Design Automationof MCMs and SiPs

Physcial Design Automationof MCMs and SiPs

Page 19: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

1904/19/23

MCM and SiP

• Multi-Chip Module• System in package (SiP)

– Different package styles– Thermal consideration for 3D

• Alternative packaging approach for high performance systems.

• Similar to PCB and IC layout problems, but– PCB layout tools cannot handle the dense and complex wiring

structure of MCM.– IC layout tools cannot handle the complex electrical, thermal

and geometrical constraints.

Page 20: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

2004/19/23

Example: Pentium

Substrate size:Substrate size:32mmx32mm32mmx32mm

Package size:Package size:43mmx43mm43mmx43mm(4 times smaller)(4 times smaller)

Page 21: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

2104/19/23

Partitioning

• Partitioning a circuit so that each sub-circuit can be implemented into a chip.

• MCM may contain as many as 100 chips.• Need to consider timing constraints and thermal

constraints• In addition, also need to consider traditional I/O

constraints and area constraints.

Page 22: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

2204/19/23

Placement

• # of components is much less as compared to IC placement.

• However, need to consider timing constraints and thermal constraints (as bare chips are placed close to each other).

• Routing is done in routing layers, not between chips.• So no routing region needs to be allocated.

Page 23: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

2304/19/23

Routing

• Main objective is to satisfy timing constraints.• Another objective is to minimize # of routing layers, not to

minimize routing area.– Cost is directly proportional to # of layers

• Crosstalk, skin effect and parasitic effect are important considerations.

• Wires are of smaller pitch and more dense than PCB layout.

Page 24: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

04/19/23 24

EE382 V -- Conclusions

Page 25: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

2504/19/23

What Have Been Taught?

• Introduced different problems in Physical Design.• Numerous algorithms which are different in terms of

– design styles– objectives– constraints– techniques– optimality– efficiency– robustness– .....

Page 26: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

2604/19/23

What Is Important?• Understand the problems

– How to formulate the problems, represent the constraints, solutions, etc.

– Reasonable assumptions/abstractions

• Know fundamental algorithms to solve the problems.• However, the world keeps on changing:

– technology– objectives– constraints– requirement on solution quality– computational power

• It is more important to learn how to think – formulate the problem– solve it smartly

Page 27: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

2704/19/23

Problem Solving Techniques

• Greedy Algorithm• Simulated Annealing/Genetic Algorithm• Mathematical Programming

– Linear, Quadratic, Integer Linear, geometric, posynomial, …

• Dynamic Programming• Reduction to graph problems

– min-cut, max-cut, shortest path, longest path, bipartite matching, minimum spanning tree, etc.

• Divide-and-Conquer• Many different heuristics• ....

Page 28: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

2804/19/23

System SpecificationSystem Specification

Micro-ArchitecturalMicro-ArchitecturalSpecificationSpecification

Timing & RelationshipTiming & RelationshipBetween UnitsBetween Units

RTL (in HDL)RTL (in HDL)

NetlistNetlist

Architectural Design

Functional Design

Logic Design

Circuit Design

VLSI Design CycleVLSI Design Cycle

Page 29: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

2904/19/23

NetlistNetlist

LayoutLayout

MaskMask

Packaged ChipsPackaged Chips

Physical Design

Fabrication

PackagingAnd Testing

VLSI Design CycleVLSI Design Cycle

Page 30: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

3004/19/23

Conventional Physical Design Cycle

PartitioningPartitioning

Floorplanning & PlacementFloorplanning & Placement

RoutingRouting

Page 31: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

3104/19/23

Technology Trend and Challenges

Source:ITRS’03

Interconnect determines the overall performanceInterconnect determines the overall performance In addition: noise, power => In addition: noise, power => Design closureDesign closure Furthermore: manufacturability => Furthermore: manufacturability => Manufacturing closureManufacturing closure

Page 32: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

3204/19/23

New Trends in Physical Design

• For nanometer IC designs, interconnect dominates• New physical effects

– Noise: coupling, P/G noise– Power: leakage, power/voltage islands– Manufacturability: yield, printability– Reliability, …

• More and more vertical integration– Logic synthesis coupled with physical design– Interconnect optimizations & design planning– Physical design as a bridge between lower level modeling

and higher level optimization/planning

• Existing CAD algorithms are far away from optimal

Page 33: 9/11/2015 1 VLSI Physical Design Automation Prof. David Pan dpan@ece.utexas.edu Office: ACES 5.434 Misc. Topics and Conclusion.

3304/19/23

Check points

Problem solving skills on underlying physical design algorithms

Know what’s behind the scene of CAD tools Know the trend and critique ability if given a new

research paper Project study of a topic of your choice and

implementation (through class project) Presentation skill Paper writing and job preparation