A High-Performance Droplet A High Performance Droplet Routing - PowerPoint PPT Presentation

A High-Performance Droplet A High Performance Droplet Routing Algorithm for Digital Mi Microfluidic Biochips fl idi Bi hi Minsik Cho and David Z. Pan Mi ik Ch d D id Z P Dept. of Electrical and Computer Engineering The University of Texas at Austin thyeros@cerc.utexas.edu dpan@ece.utexas.edu http://www.cerc.utexas.edu/utda Sponsored in Part by NSF and IBM Faculty Award 1

Why Biochip? KCL Bovine Tris-HCL Gelatin MgCl 2 Serum Albumin Mix Mix Mix Mix D Deoxy- nucleotide Primer Triphosphate Mix Mix Human Experimenter λ DNA AmpliTaq DNA p q Polymerase Mix Mix Mix Result DNA PCR Biochip Biochip � Economical and high-performance › Low cost (less than $2), portable, disposable ( $ ), p , p › Fast, automated, error-tolerant (no human involvement) � Critical applications › POC (Point-of-care), anti-bioterrorism, … 2

Digital Microfluidic Biochips [Courtesy Advanced Liquid Logic] Ctrl Ckt Ctrl Ckt Preprogrammed voltages for EWOD P d lt f EWOD [IEMN] � Digitized droplets transported Di iti d d l t t t d › By EWOD (electrowetting-on-dielectric) y ( g ) » Electrical modulation of the solid-liquid interfacial tension › According to Preprogrammed Schedule p g g » Traffic control 3

Microfluidic Biochip Droplet Routing � How to program/schedule/route droplets? › Fluidic constraints to prevent collision (keep-off distance) p ( p ) › Timing constraints to prevent spoilage › NP-Complete 9 T 1 S 2 T 1 S 2 5 1 4 1 6 7 8 3 X X 2 2 T 2 2 3 3 2 2 2 T 7 2 2 T 2 T 1 5 6 1 4 3 S 1 S 1 S S 1 [Courtesy Advanced Liquid Logic] T=?? (infeasible) T=9 (good solution) � Comparison with VLSI routing � Comparison with VLSI routing › Fluidic constraint = Minimum spacing › Timing constraint = Required arrival time (RAT) › Timing constraint Required arrival time (RAT) › But, time-multiplexed movement of droplets 4

Modeling and Constraints t y x x,y,t +1 x,y,t +1 t+1 1 x +1 ,y,t +1 x,y +1 ,t +1 x,y,t x,y -1 ,t +1 x -1 ,y,t +1 ( x,y,t ) y+1 y Graph model for simultaneous p t-1 1 geometric and temporal scheduling y-1 x-1 x+1 � Graph model › 5 edges for each node › Time causality � Fluidic cube › One droplet inside the cube 5

Current State-of-the Art � Prioritized A* search [Böhringer TCAD’06] › Route shorter droplets first (widely used in VLSI) � Network flow-based approach [Yuh+ ICCAD’07] › Maximize the number of nets routed › Min cost-Max flow formulation + prioritized A* search � OSPF protocol approach [Griffith+ TCAD’06] › Have a set of precomputed path, and choose one of them by H t f t d th d h f th b situation based on OSPF network protocol. � Two-stage Algorithm [Su+ DATE’06] � Two stage Algorithm [Su DATE 06] › Generate M shortest paths › Random selection � Progressive ILP based Approach [Yuh+ DAC’08] › Similar to VLSI routing › Pin constraints 6

Proposed Approach � Time-multiplex resource sharing implies › Intermediate paths will be freed up eventually. � To reduce problem size inspired by Chatin’s coloring algorithm � New concepts to reflect the nature of biochip › One droplet movement at a time (the others are frozen) » Reduced routing search time › Bypassibility » To route a droplet with minimal impact on feasibility T t d l t ith i i l i t f ibilit › Concession » To resolve a deadlock » To resolve a deadlock › Compaction » To satisfy timing constraint and improve fault-tolerance 7

Overall Routing Flow success Start failure yes Routing by Routing by deadlock? Bypassibility Concession failure failure no success yes yes no no Greedy Greedy Unroutable routed? optimization � Reduce the problem size › To find out the most complex part of the problem › To find out the most complex part of the problem � First find a feasible solution › Greedily improve the solution to meet timing › Greedily improve the solution to meet timing 8

T j T j T j T i T i T i T i S k S k S k S i S i S i T k T k T k T k S j S j S j One Droplet at a Time k i j 9 x t y y t 3 t 3 t 2 t 1 t 0

Routing by Bypassibility 9 T 1 S 2 4 4 3 3 8 8 2 2 1 1 T 5 7 T 2 V left V right 1 1 4 4 5 5 6 6 2 2 3 3 S 1 T=9 (optimal solution) H down Full bypassibility Full bypassibility No bypassibility � A routed droplet will block the target regions. › Is there any H/V bypass for the unrouted droplets? I th H/V b f th t d d l t ? � Four categories › Ideal : the target is a waste reservoir › Ideal : the target is a waste reservoir. › Full : both horizontal and vertical bypasses are available. › Half : only either horizontal or vertical bypasses is available. › No : no bypass is available. 10

Routing by Concession � When there is a deadlock… › One droplet needs to back-off O d l t d t b k ff › One closer to the empty space p y p 11

S 5 S 4 S T 6 Toy Example (Bypassibility) T 3 T 3 12 S 6 S T 4 T 2 T 5 T 1 S 3 S 1 S 1 S 2 T S

S 5 S 4 S 4 S S T 6 T 6 Toy Example (Bypassibility) T 3 T 3 T 3 T 3 13 S 6 S 6 S S T 4 T 2 T 5 T 1 S 3 S 1 S 1 S 2 T S

S 5 S 4 S T 6 Toy Example (Concession) T 3 T 3 T 3 T 3 14 S 6 S T 4 C-Zone C Zone T 2 T 5 T 1 S 3 S 1 S 1 S 2 T S

S 5 S 4 S T 6 Toy Example (Bypassibility) Operation time = 72 T 3 T 3 15 S 6 S T 4 T 2 T 5 T 1 S 3 S 1 S 1 S 2 T S

S 5 S 4 S T 6 Toy Example (Greedy Opt.) Operation time = 19 T 3 T 3 16 S 6 S T 4 T 2 T 5 T 1 S 3 S 1 S 1 S 2 T S

Experimental Setup � Implemented in C++ and tested on Intel Dual Core 2.6GHz Linux with 4GB � Two benchmarks › Suite1: 4 widely used benchmarks y » Relatively small and easy » Max # of droplets: 6 › Suite2: 30 synthetic benchmarks S it 2 30 th ti b h k » Large scale and complex with multiple blockages » Max # of droplets: 48 with 30% of areas blocked » Max # of droplets: 48 with 30% of areas blocked � Comparison › Prioritized A* search [TCAD’06] › Prioritized A search [TCAD 06] › Two-stage algorithm [DATE’06] › Network flow based routing [ICCAD’07] g [ ] 17

Comparison betw een Suite1 and Suite2 [Yuh+ ICCAD’07] Much More Complex For Future Design For Future Design 18

Experimental Results 4 P-A* [TCAD'06] P-A [TCAD 06] Two-Stage[DATE'06] NetFlow [ICCAD'07] 3 Ours O Number of 2 failed designs 1 0 0 Suite1 � Total 4 designs � Completed the most number of designs � Completed the most number of designs 19

Experimental Results 25 P A* [TCAD'06] P-A* [TCAD 06] NetFlow [ICCAD'07] 20 Ours Ours Number of 15 failed designs 10 10 5 3 0 Suite2 � Total 30 designs � Completed the most number of designs � Completed the most number of designs 20

Experimental Results 120 P A* [TCAD'06] P-A* [TCAD 06] 100 NetFlow [ICCAD'07] Ours Ours 80 80 Number of 60 the unrouted droplets 40 3 20 0 Suite2 � Total 864 droplets (30 designs) � Outperforms the previous works by wide margin � Outperforms the previous works by wide margin 21

Conclusion � New digital microfluidic biochip routing algorithm › Multiple concepts to leverage time multiplexed › Multiple concepts to leverage time multiplexed resource sharing � Outperforms the previous works by wide margin � Outperforms the previous works by wide margin � Ping-Hung Yuh, Prof. Chia-Lin Yang, and Prof. Yao Wen Chang (NTU) Yao-Wen Chang (NTU) › Providing the results of network flow-based on algorithm in ICCAD’07 on benchmark suite2 algorithm in ICCAD 07 on benchmark suite2 Thank you!! Thank you!! 22