NCKU CSIE EDALAB

Kai-Han Tseng, Sheng-Chi You, Jhe-Yu Liou and Tsung-Yi Ho

http://eda.csie.ncku.edu.twDepartment of Computer Science and Information Engineering

National Cheng Kung UniversityTainan, Taiwan

2013 ACM International Symposium on Physical Design

NCKU CSIE EDALAB

Outline

Introduction

Problem Formulation

Algorithms

Experimental Results

Conclusions

NCKU CSIE EDALAB

What is Biochips?

․General definition

A chip with a small solid platform made of glass, plastic, or

membrane which deals with the behavior, precise control and

manipulation of fluids that are constrained to a small scale

․Functionality

Analysis, reaction, or detection of biological samples (DNA or

human blood)

․Application Clinical diagnostics

Environmental monitoring

Automated drug discovery

Cell culturing

Protein crystallization

NCKU CSIE EDALAB

Classification of Biochips

Biochips

Microfluidic biochipsMicroarray

DNA chip Protein chipDroplet-based

Continuous-flow

Flow-based microfluidic biochips

Flow-channel

Valve

NCKU CSIE EDALAB

․Functional component

Valves can be combined to form more complex units,

e.g., latches, switches, mixers, multiplexers, micropumps

․Reliability

The open and close activities (valve-switching) for a valve is

limited

Valve: The Basic Unit of Flow-based Microfluidic Biochips

NCKU CSIE EDALAB

The Need of CAD Support

․Applications become more complicated

Large-scale bioassays

Multiple and concurrent assay operations on a biochip

․Design complexity is increased

The increasing rate of the valve numbers is four times faster

than Moore’s Law

․Current methodologies

Manual

Full-custom

Source: Fluidigm

A top-down synthesis methodology is needed

NCKU CSIE EDALAB

Component Model : Mixer

Microfluidic mixer

S2S1 S3

Pump

NCKU CSIE EDALAB

0 0

Phase V1 V2 V3 V4 V5 V6 V7 V8 V9

1. Ip1

2. Ip2

3. Mix

4. Op1

5. Op2

1 1

Motivation

Microfluidic mixer

Open the valve

Close the valve

0 0 0 0 0 0 0

0 0 0 0 0 0 00 0

0 0 0 0 0 00 0 0

0 0 0 0 00 0 0 0

0 0 0 0 0 0 0 0 0

1 1

1 1

1 1

1 1

123

5 6

78

9

4

MixMixMix

Valve-switching : 00020610141820 + Mixing

Out2Out2

In1 Out1In2 Out2Mix

20

In1 Out1In2 Out2Mix

20 20

Reduce valve-switching 6 times

Valve-switching for a mixing operation

Total reduction

In1

14 6

146

In1 Out1In2 Out2Mix

20

5 Phases for a mixing operation

In1 Out1In2 Mix

Out1In2 MixIn1

In1 Out1In2 Out2Mix20

20

Out2

12

Operation1

Operation2

Operation3

Out2

In1 Out1In2 Out2MixIn1

208

146

6

24

Operation1

Operation2

Operation3

Reduce valve-switching and the application time

Group Continuous Operations

NCKU CSIE EDALAB

․Total valve-switching

Valve-switching of a mixer

Valve-switching of intersections

Valve-switching of a storageStorage

Transportation Issue

NCKU CSIE EDALAB

Outline

Introduction

Problem Formulation

Algorithms

Experimental Results

Conclusions

NCKU CSIE EDALAB

Component Phases

Mixer Ip1 / Ip2 / Mix / Op1 / Op2

Filter Ip / Filter / Op1 / Op2

Detector Ip / Detect / Op

Separator Ip1 / Ip2 / Separat/ Op1 / Op2

Heater Ip / Heat / Op

Problem Formulation

․Input: A biochemical application modeled as a

sequential graph and a component library

․Objective: Synthesize a flow-based microfluidic biochip

such that the valve-switching amount and application

completion time are minimized

․Constraint: Resource constraints and design rules

Parameter Value

Width of flow channel 100 μm

Minimum spacing between flow channels 40 μm

Width of control channel 30 μm

Width of control valve 100 μm

Minimum spacing between control channels 40 μm

Minimum spacing between external ports 1500 μm

NCKU CSIE EDALAB

Outline

Introduction

Problem Formulation

Algorithms

Experimental Results

Conclusions

NCKU CSIE EDALAB

Input

Application

Graph

Component

Library

Mixer : 3

Resource

Constraint

NCKU CSIE EDALAB

Top-Down Synthesis Methodology

․Architecture-level synthesis

Set-based minimum cost

maximum flow (SMCMF)

Resource binding

Resource scheduling

․Physical-level synthesis

Incremental cluster expansion

(ICE)

Placement

Routing

․Explicit scheduling

List-scheduling based

approach

Total valve-switching amount

Application completion time

NCKU CSIE EDALAB

Set-based Minimum Cost Maximum Flow

․Set-based

Group continuous operations in a set to reduce the valve-

switching activities

․Maximum Flow

Each flow path represents a component. Our goal here is to

maximize the component parallelization.

․Minimum Cost

Find a way that not only satisfies the parallelization but also

minimize the application completion time

NCKU CSIE EDALAB

Time Evaluation

Transportation time : 3

NCKU CSIE EDALAB

2325

1013

1620

1620

1013

1013

17

1625

2831

513

02

02

57

Find Continuous Sets

O1 O2 O3 O4 O5 O6 O7

O8O9 O10

O11 O12

O13

NCKU CSIE EDALAB

O1

O8

O11

O13

0

31

S1

O2

0

2

S2

O3

O9

5

13

S3

O4

1

7

S4

O5

O10

O12

10

25

S5

O6

10

13

S6

O7

16

20

S7

S : Estimated starting timeE : Estimated ending time

Si

SE

Continuous Sets

Build the Flow Network

0

31

0

2

5

13

1

7

10

25

Src

V1

V5

․ Step1

Create Src and Dst vertex

Create a vertex Vi ϵ V for each Si

․ Step2

Separate Vi → (Viin, V

iout)

Create a directed edge Viin → Vi

out,

cost=0 and capacity=1

․ Step3

Separate Dst → (Dstin, Dstout)

Create a directed edge Dstin → Dstout,

cost=0 and capacity=3 (given

component numbers)

․ Step4

Create directed edges Src → Viin and

Viout → Dstin, cost = 0 and capacity=1

For each Viout and Vj

in, if Viout + T

(transportation time) ≦ Vjin, create a

directed edge Viout → Vj

in, cost=f(x)

and capacity=1. Here, x is defined as

|(Viout -Vi

in)| and f is proportional to x

V1in

V2in

V3in

V4in

V7in

V6in V

5in

V3

out

V1out V

7out

V6

out

V5

out

V4

outV2

out

DSTin

DSTOUT

V2

V3

V7

V6

V4

Dst

10

13

16

20

NCKU CSIE EDALAB

Apply Minimum Cost Maximum Flow

Src

V1

V5

V2

V3

V4

V6

Dst

V7

V6 is isolated

Find 4 concatenated-sets

NCKU CSIE EDALAB

Sort by the Priority

O1

O8

O11

O13

O6

O3

O9

O2

O7

O5

O10

O12

O4

Mixer1 Mixer2 Mixer3

2 continuousoperations

3 continuousoperations

high low

Priority 1. Number of continuous operations2. Number of total operations

The remaining operation

NCKU CSIE EDALAB

Next operation

Insert the Remaining Operation

O3

O9

O2

O7

Mixer3

O1

O8

O11

O13

Mixer1 Mixer2

O5

O10

O12

O4

O6

Mixer2

O10

O12

O6

O4

O5

O1

O8

O11

O13

Mixer1 Mixer3

O2

O7

O3

O9

NCKU CSIE EDALAB

Relational Graph

M1 M3

M2

1

11 2

2

Relational graph

M3 -> M1

NCKU CSIE EDALAB

Top-Down Synthesis Methodology

․Architecture-level synthesis

Set-based minimum cost

maximum flow (SMCMF)

Resource binding

Resource scheduling

․Physical-level synthesis

Incremental cluster expansion

(ICE)

Placement

Routing

․Explicit scheduling

List-scheduling based

approach

Application completion time

Total valve-switching amount

NCKU CSIE EDALAB

Motivation

User User

User

User

User

User

User

UserUser

User

User

User

User

User

User

User

User

User

User

User

User

User

UserUserUser

User

User

User

User

User

User

User

M2

S1 M3

M4

M1

M5

S1 S2 M5

M4

M3 M1

M2

O(n2)

NCKU CSIE EDALAB

One Storage

Storage1

4*2 + 2Valve-switching

NCKU CSIE EDALAB

Two Storage

Storage1 Storage2

3*2 + 2Valve-switching

NCKU CSIE EDALAB

Find the Clusters

M10

M9

M8

M7

M4

M1

M5

M2

M3

M6

Iteration :Cluster :

2

1

2

3

2

3

51

1

4

4

4

1

1

1

01019786

1

Storage : 4

NCKU CSIE EDALAB

Find the Clusters

M10

M9

M8

M7

M4

M1

M5

M2

M3

M6

2

1

2

3

2

3

1

1

1

1

1

Iteration : Cluster :

2654

1

Storage : 4

NCKU CSIE EDALAB

Incremental Cluster Expansion

M3

M4 M6

M2

M7

M8

S4 S3

M10

M1S1 S2

M5M9

NCKU CSIE EDALAB

Compaction

S1M1

M10 M5

M2

M9

S2

M7

M8

S4M3

M4 M6

S3

NCKU CSIE EDALAB

Routing by dijkstra’s algorithm

M5

M2

M9

S2

M3

M4

S3

S1

M10

M7

M8

S4

M1

M6

NCKU CSIE EDALAB

100 Mixers and 10 Storages

NCKU CSIE EDALAB

Top-Down Synthesis Methodology

․Architecture-level synthesis

Set-based minimum cost

maximum flow (SMCMF)

Resource binding

Resource scheduling

․Physical-level synthesis

Incremental cluster expansion

(ICE)

Placement

Routing

․Explicit scheduling

List-scheduling based

approach

Application completion time

Total valve-switching amount

NCKU CSIE EDALAB

Time 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35

Mixer1

Mixer2

Mixer3

Time 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35

Mixer1

Mixer2

Mixer3

Valve-switching amount : 250 (62,188)

Application Completion Time : 28.5(s)

Color / Work

Operation Trasportation

Idle Input Output

Op1

Op4

Op2

Op8

Op3 Op9

Op5

Op11

Op7

Op6 Op10 Op12

Op13

Valve-switching amount :390(166,224)

Application Completion Time : 33.2(s)

Op13Op1

Op2

Op4

Op8

Op3

Op6

Op5

Op7

Op9

Op11

Op10

Op12

Reduce : 4.7(s)

Binding by SMCMF

Binding by baseline method

Explicit Scheduling

Reduce 140 valve-switching

NCKU CSIE EDALAB

Outline

Introduction

Problem Formulation

Algorithms

Experimental Results

Conclusions

NCKU CSIE EDALAB

Experimental Settings

․Implemented our algorithm in C++ language on a PC with Core2 Quad processors at 2.66GHz and 3.25GB of RAM

․Compared set-based minimum cost maximum flow binding algorithm with list scheduling based binding algorithm (baseline method) on top-down synthesis methodology

․Tested on several synthetic benchmarks by Adjusted operation numbers from 7 to 8191 and fixed mixers as

20 and storage as 1

Adjusted mixers from 10 ~ 100 and fixed operation numbers as 1023 and storage as 1

Adjusted storage from 1 ~ 10 and fixed operation numbers as 1023 and mixer as 100

NCKU CSIE EDALAB

Experimental Results (1/3)

Application Completion Time Valve-switching Amount

Operation : 7~8191

Mixer : 20Storage : 1

NCKU CSIE EDALAB

Experimental Results (2/3)

Operation : 1023Mixer : 10 ~ 100Storage : 1

Valve-switching AmountApplication Completion Time

NCKU CSIE EDALAB

Experimental Results (3/3)

Valve-switching Amount

Operation : 1023Mixer : 100Storage : 1 ~ 10

Application Completion Time

NCKU CSIE EDALAB

Outline

Introduction

Problem Formulation

Algorithms

Experimental Results

Conclusions

NCKU CSIE EDALAB

Conclusions

․The valve-switching activities for the components such

as mixer and storage were modeled

․A first top-down synthesis methodology for flow-based

microfluidic biochip was presented

․Two algorithms for architecture-level synthesis and

physical-level synthesis were proposed

Set-based minimum cost maximum flow (SMCMF)

Incremental cluster expansion (ICE)

․The experimental results showed that our top-down

synthesis methodology not only minimized the valve-

switching amount but also reduced the application

completion time

NCKU CSIE EDALAB