Manufacturing of Lab-on-a-Chip Devices: Variation Analysis of Liquid Delivery using Blister Packs by Sivesh Selvakumar B.E. Mechanical Engineering (2009) College of Engineering, Guindy, India Submitted to the Department of Mechanical Engineering in Partial Fulfillment of the Requirements for the Degree of Master of Engineering in Manufacturing at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY Massachusetts Institute of Technology NOV 0 4 2010 September 2010 LIBRARI ES © 2010 Sivesh Selvakumar All rights reserved ARCHIVES The author hereby grants to MIT permission to reproduce and to distribute publicly paper and electronic copies of the thesis document in whole or in part in any medium now known or hereafter created Signature of Author............ .......... e artm t of Mechanical Engineering .. g.. ..'21ust 2 " 10 ....... . ....... Certified by........... A..h. .... Lectu 71 epartment of Mecanical Engineering A -"Ii ThesMwSupervisor A ccepted by............................................ David E. Hardt Chairman, Committee on Graduate Students This page has been intentionally left blank Manufacturing of Lab-on-a-Chip Devices: Variation Analysis of Liquid Delivery using Blister Packs by Sivesh Selvakumar B.E. Mechanical Engineering (2009) College of Engineering, Guindy, India Submitted to the Department of Mechanical Engineering in Partial Fulfillment of the Requirements for the Degree of Master of Engineering in Manufacturing Abstract Components for on-chip storage and delivery of liquid reagent are necessary for many commercial applications of lab-on-a-chip technology. One such system uses a 'blister pack' that is pushed by an actuator. Over the course of product development, Daktari Diagnostics had completed nominal design of a blister pack for their flow rate requirements. This work involved performing a thorough variation analysis of the blister pack to determine the critical sources of variation. For this purpose, the tool of variation simulation modeling (VSM) was used. A numerical model of the blister performance was developed and Monte Carlo simulations were conducted. The results showed that this fluid delivery technique is robust and the number of out-of-specification parts was less than 2%. The critical blister pack parameters that must be controlled were also determined and these can be used to improve process capability. Thesis Supervisor: Dr. Brian Anthony Title: Lecturer 4 Acknowledgements I wise to convey my gratitude to: Aaron, Bill, Robert, Peter,Andy and Betsy, for being such great hosts at Daktari Diagnostics. Your enthusiasm and drive for the tasks at hand never ceased to amaze me. Working for Daktari has been an absolute pleasure. Rodrigo, my team-mate and friend, for his motivation and for the 'occasional' advice. Much of this work owes its existence to your input and hard work. It has been a pleasure to work with you. Dr. BrianAnthony, our thesis advisor, for his support and encouragement throughout what has been, in many ways, a unique project experience. Jennifer Craig, our writing advisor, who surely spent many hours reading several last-minute drafts. Without her support and eye for detail, this thesis would have suffered in quality. The folks at the Edgerton Student Shop for their understanding and much-needed suggestions through many hours of machining. This project has gained a lot from their experience. Amma, Appa and Appayya for making all of this possible. 6 Table of Contents Acknowledgements 5 Table of Contents 7 List of Figures 11 List of Tables 13 1. Introduction 15 1.1 The M.Eng. Capstone Project 16 1.2 Overview of the Thesis 16 2. Product and Project Overview 19 2.1 Company Background 19 2.2 ProductDescription 19 2.2.1 The Cartridge 20 2.2.2 The Instrument 21 2.2.3 Product Timeline 22 2.3 Problem Statement 23 2.3.1 Survey of Manufacturing Issues 23 2.3.2 Selection of Subset of Problems 25 3. Background Research 3.1 Lab-on-a-ChipApplications 3.1.1 CD4+ Cell counting through Cell Lysate Impedance Spectroscopy 3.2 MicrofluidicDevice Architecture 27 27 27 28 3.2.1 Central Layer 29 3.2.2 External Layer 29 3.2.3 Additional Components 29 3.3 Lab On a Chip Technologies 30 3.4 Lab-on-a-Chip ManufacturingProcesses 31 3.4.1 Part Manufacturing 31 3.4.2 Functionalization Process 32 3.5 Monte CarloAnalysis 33 3.6 Machine Vision for Metrology 34 3.6.1 Choice of Metrology System 34 5.1.2 Components of a Machine Vision System 34 4. Development of the Blister Performance Model 37 4.1 Working of the DaktariBlister Pack 37 4.2 Requirements of the Blister Numerical Model 38 4.2.1 Blister Outputs and Noise Factors 4.3 The Numerical Model 40 41 4.3.1 Input Blister Dimensions and Speed of Travel 43 4.3.2 Step Size 43 4.3.3 Determine Actuator Positions according to Step Size and Misalignment 44 4.3.4 Calculate the total volume of fluid expelled up to the current step 46 4.3.5 Volume expelled in the current step 51 4.3.6 Getting the 'instantaneous' flow rate 51 4.3.7 Calculate the average, maximum and minimum flow rates 53 4.4 Assumptions 5. Validation of the Blister Model 5.1 MeasurementSetupfor Blister Dimensions 53 55 55 5.1.2 Selection of Hardware 55 5.1.3 Selection of Software 56 18.104.22.168 Gauge Repeatability and Reproducibility 57 5.2 MeasurementSetupfor Blister Flow 58 5.3 Validation of Blister Numerical Model 58 6. Monte Carlo Analysis and Regression Models 61 6.1 Manufacturing Variation Data 61 6.2 Monte CarloAlgorithm 62 7. Results and Discussion 63 7.1 Flow Rate Variation after considering only BlisterDimensional Variation 63 7.2 Flow Rate Variation after considering the variation in all dimensions 65 7.3 Regression on the Average Flow Rate 67 7.4 Regression on the Range of Flow Rates 68 7.5 Effect of Variation in Individual Dimensions on Flow Rate 69 7.5.2 Effect of Spherical Radius Variation 69 7.5.3 Effect of Blister Height Variation 70 7.5.4 Effect of Actuator Radius Variation 71 7.5.5 Effect of Starting Height Variation 72 7.6 Flow Rate Variation after consideringthe effect ofshimming 73 8. Conclusions and Recommendations 75 9. Future Work 77 9.1 BlisterModel Validation 77 9.2 Study of Dimensional Variation in Blisters 77 9.3 Increasing the Number of Runs in the Monte Carlo Simulation 77 9.4 Electrode Foils - ConfigurationStudy and ProcessAnalysis 77 9.5 FunctionalizationProcessOptimization 78 9.6 Injection Molding of the Backbone - Normalization Time Optimization 78 9.7 Valve-Solenoid Interaction Robustness Study 78 9.8 Effect of Imprecise Actuator Movement 78 References 81 10 List of Figures Figure 1: Daktari Diagnostics CD4 Cell Counting Platform 19 Figure 2: The Daktari cartridge - with parts marked 20 Figure 3: The Daktari instrument - with parts marked 21 Figure 4. Assay Process Diagram 28 Figure 5. Microfluidic Device Architecture 29 Figure 6. Functionalization Process 33 Figure 7: Schematic of Metrology System using Machine Vision 35 Figure 8: Photograph of a Blister Pack (above) and Actuators (below) 37 Figure 9: Blister actuation process 38 Figure 10: Required Outputs from a Mathematical Model of the Blister 39 Figure 11: Particular blister dimensions that affect the flow rate 40 Figure 12: Flowchart of Blister Numerical Model 42 Figure 13: Effect of small step size on flow rate measurement 43 Figure 14: Effect of large step size on flow rate measurement 44 Figure 15: Introducing the effect of actuator misalignment and step size 44 Figure 16: Usage of transformation matrices to convert actuator coordinates to the blister coordinate system 45 Figure 17: Discrete Volume Calculation 46 Figure 18: Flowchart for Blister Volume Calculation 47 Figure 19: Intersection of Actuator and Blister at a particular height 48 Figure 20: Possible Scenarios when a Blister and Actuator Intersect 49 Figure 21: Calculating the area of a Circle-Circle Intersection 50 Figure 22: Volume Expelled in the Current Step 51 Figure 23: Total Volume Expelled vs. Crush Depth 52 Figure 24: Time taken for a Step versus the Crush Depth 53 Figure 25: Schematic of Blister & Camera Setup 55 Figure 26: Photograph of Blister Measurement Setup 56 Figure 27: Photographs of the Blister Using the Current Setup 56 Figure 28: Blister diameter (in no. of pixels) measured by 2 operators 57 Figure 29: Blister Height (in no. of pixels) measured by 2 operators 57 Figure 30: Plot of Experimental and Predicted Flow Rates from 5 Blisters 59 Figure 31: Flowchart of the Monte Carlo Algorithm 62 Figure 32: Distribution of Flow Rates considering only Blister Variation 64 Figure 33: Distribution of Flow Rates considering variation in all dimensions 66 Figure 34: Flow Rate Profile with a Smaller Spherical Radius 69 Figure 35: Flow Rate Profile with a Larger Spherical Radius 70 Figure 36: Flow Rate Profile with a Smaller Blister Height 70 Figure 37: Flow Rate Profile with a Larger Blister Height 71 Figure 38: Flow Rate Profile with a Smaller Actuator Radius 71 Figure 39:Flow Rate Profile with a Larger Actuator Radius 72 Figure 40: Flow Rate Profile with Lower Actuator Starting Height 72 Figure 41: Flow Rate Profile with Higher Actuator Starting Height 73 Figure 42: Distribution of Flow Rates considering the effect of shimming 74 List of Tables Table 1: Product Development timeline 22 Table 2: List of Potential Cartridge-Instrument Interaction Issues 25 Table 3: List of Potential Issues during Ramp-up 26 Table 5: Variation Data for the Blister and Actuator Dimensions 61 Table 6: Summary of Output Distribution Properties considering the variation in blister dimensions alone63 Table 7: Increase in Percentage of Non-conforming Blisters with tighter tolerance ranges 65 Table 8: Summary of Output Distribution Properties considering the variation in all relevant dimensions 67 Table 9: Increase in Percentage of Non-conforming Blister-Actuator Systems with tighter tolerance ranges 67 Table 10: List of Coefficients for Linear Regression on Average Flow Rate 68 Table 11: List of Coefficients for Linear Regression on Range of Flow Rates 68 Table 12: Summary of Output Distribution properties considering the variation in all relevant dimensions73 Table 13: Percentage of Non-conforming Blister-Actuator Systems after shimming 74 14 1. Introduction The field of microfluidics has been the subject of several decades of research. As a sub-field of MEMS/MST (Micro Electromechanical Systems/ Microsystems Technology), microfluidics promises the integration of various liquid preparation/processing steps along with the sensors (optical, electronic or otherwise) that enable the complete analysis of samples. These so-called labs-on-a-chip (LOC) have the potential to revolutionize any application where the analysis or use of small quantities of fluid is useful. Some of these applications include immunoassays, pharmaceutical drug discovery, genomics and proteomics, cytology and biotechnology, drug delivery and surface patterning . Despite this immense potential, till date, the commercial impact of the field has been minimal . This may be attributed to several factors: the absence of commercial needs, low understanding of fabrication technology and the lack of expertise in microfluidic product industrialization. However, the situation appears to be changing. Point-of-care (POC) diagnostics has recently emerged as an ideal market for LOC technology. Currently, most diagnostic tests are performed in centralized laboratories which account for around 60% of the 2004 market revenue . Another 30% of the market is commanded by POC diagnostics but these are almost exclusively for glucose-monitoring. For a number of other pressing problems such as HIV, TB and malaria, POC solutions are practically non-existent. Microfluidic technology can make these solutions possible . The size of the problems that these diseases pose can be understood by the sheer number of people affected by them. HIV affects nearly 33 million people worldwide . TB affects another 9.4 million causing around 1.3 million deaths . Nearly 300-500 million people are infected with malaria . For all these infectious diseases, POC diagnostics promises to decrease the time between the first visit to the doctor and the beginning of treatment . In many situations, this can lead to improved health outcomes. For example, for HIV patients on anti-retroviral therapy (ART) a key test, called the CD4 count, is used to monitor their immune system . Recent efforts to scale ART in developing countries quickly highlighted the CD4 test as a significant bottleneck. POC promises to relieve this problem. Keeping pace with the growing interest in LOC, is the constantly increasing understanding of microfluidic fabrication technology. A great deal of work has been done [9-13] to better understand and optimize these processes. Several organizations (e.g.: Inverness, Abbott Point of Care, Daktari Diagnostics, Claros, Diagnostics 4 All) are taking advantage of these developments and are actively involved in the POC equipment field. Thus, there is a growing interest in the process of microfluidic product industrialization. As a new entrant in the medical device space, Daktari Diagnostics (hereafter referred to as Daktari) is currently involved in the process of product industrialization. A number of issues related to reliable production of their microfluidic product are being addressed during this phase. Some of these issues are: product tolerance analyses, design for manufacture and assembly (DFMA) and identification of critical-to-function paths. In this thesis, the manufacturability of one component of the Daktari LOC system - the 'blister pack' - has been studied extensively. The blister pack is responsible for storing liquid reagents in the product and then delivering these at a controlled flow rate. Its performance is affected in many ways by manufacturing variation and this effect was fully explored. Numerical models relating blister performance to their dimensions were developed and key input parameters that must be controlled to ensure good performance were identified. 1.1 The M.Eng. Capstone Project This thesis is the result of a group project that has been completed as part of the requirements of the Masters of Engineering in Manufacturing program at the Laboratory for Manufacturing and Productivity (LMP) at MIT. A team of students approached the research problems collaboratively and then each team member focused on a different research challenge. The author of this work, Sivesh Selvakumar, focused on blister manufacturing issues. Rodrigo Linares  focused on instrument manufacturing issues. Together, the two theses represent the complete analysis of the problem of microfluidic device tolerance analysis. 1.2 Overview of the Thesis This thesis begins with a description of the company, product and problem statement in Chapter 2. This is followed by a comprehensive literature review of the state-of-the-art in microfluidic fabrication technology and assembly strategies in Chapter 3. Chapter 4 presents the development of a numerical model that relates the geometry of the 'blister pack' to its flow performance. Efforts to validate this model are presented in Chapter 5. Chapter 6 describes the methodology of Monte Carlo simulations used to study the effect of manufacturing variation on the blisters. The results of these simulations along with suitable conclusions are presented in Chapter 7. Our recommendations for future work follows in Chapter 8. 18 ..... .................. ... ..................... ................ 2. Product and Project Overview 2.1 Company Background Daktari Diagnostics is a medical diagnostic company located in Cambridge, Massachusetts, USA that utilizes unique technology for various high-impact diagnostic applications. The company has dedicated its product line to various diagnostic tests that will have a strong positive effect on global health. Their first product is a point-of-care diagnostic platform for HIV patients. The product is designed to be easy-to-use, robust and inexpensive - suited to developing countries where this test is a major bottleneck to effective delivery. 2.2 Product Description The diagnostic platform that is being designed by Daktari Diagnostics is a CD4 counter that is necessary for HIV patients. Similar to a glucometer (which provides patients suffering from diabetes with a value of their blood glucose concentration), the CD4 counter provides caregivers with the white blood cell concentration in an HIV patient's blood. This value indicates the relative strength of the patient's immune system and can be used to determine the appropriate dosage and strength of anti-retroviral drugs (ARV). Figure 1 shows the process that goes into a CD4+ T-Cell count using Daktari's System. 1.Blood is introduced to disposable, which is inserted into instrument 2. CD4 cells are captured chermicaly, without sample preparation Cartridge 3. CD4 cells are detected and counted electrochenically, without optics or labels Instrument Figure 1: Daktari Diagnostics CD4 Cell Counting Platform The typical use scenario involves a trained operator carrying the portable 'instrument' and a set of 'cartridges' to remote locations where the test is unavailable in its current form. Using a lancet, the operator pricks the patient's finger and allows the patient's blood to enter the card at a sample entry port. The card draws the blood in by means of capillary action. After sufficient blood is drawn in, the sample entry port is closed using a cap and the card is loaded into the instrument. The instrument begins driving the blisters in a carefully designed sequence of operations that prepares the blood sample and then sends it through the assay chamber. The antibodies deposited on the channel capture the CD4 cells. The cells are then lysed ('lysis' refers to the rupture of a cell's membrane) in a high-impedance solution. This releases low-impedance cellular contents into the assay chamber. The difference in impedance at this stage is used to determine the concentration of cells in the blood . The CD4+ cell concentration value is then displayed using an LCD display and the test concludes. The platform consists of two parts: an instrument and several disposable cartridges. 2.2.1 The Cartridge Each cartridge is a disposable microfluidic device complete with reagents and a sensing mechanism to measure the concentration of CD4 cells in blood. Figure 2 depicts a prototype of the microfluidic card with the following 6 parts: Housing Blister Cap Backbone with microfluidic channels Lid foil on the back Electrode Foil Figure 2: The Daktari cartridge - with parts marked 1. Backbone - an injection molded polymer card with microfluidic channels ck . .... .... .... .. . . ... ............ ...... ........ 2. Lidfoil - a transparent polymeric sheet that is bonded to one side of the backbone and forms the fourth wall of the channels 3. Functionalizedfoil- a polymeric foil that covers the 'assay chamber' where the CD4 cell count is performed. This foil has a gold electrode layer formed on it. It is then coated with antibody solution, which is used to immobilize the target CD4 cells. 4. Blisterpack - similar to a pharmaceutical blister pack, this part contains the liquid reagents that are necessary for sample preparation 5. Housing - an injection molded polymer part that protects the blister pack and functionalized foil during transport 6. Cap - plastic part that covers the sample entry port after blood is introduced into the cartridge 2.2.2 The Instrument The instrument is designed to be portable, battery-powered and electronically controlled. It contains linear actuators to 'pop' the blisters and flow reagent through the cartridge's channels. Figure 3 shows a photograph of the instrument. Handle for holdin the device Figure 3: The Daktari instrument - with parts marked The instrument also has suitable electronic interfaces, which connect to the cartridge and measure the impedance within the assay chamber. This impedance measurement is used to determine the final level of CD4 cells within the device. In addition, the electronics necessary to drive the actuators, calculate the CD4 count and output this value to the LCD display are also present. The instrument is both the driver of the reagents and the user interface medium between the system and the user. In order to accomplish all these functions, the proper integration of the following subassemblies must be accomplished. 1. Frame - the main structural component of the instrument. It locates the different subassemblies. 2. Door Subassembly - holds the cartridge in place and guarantees planarity between the cartridge and theframe. 3. Actuator Subassembly - carries the actuators and ensures perpendicularity between these and theframe. 4. Solenoid Subassembly - supports the valve solenoids and ensures perpendicularity between the solenoids and theframe. 5. Outer Casing - provides both aesthetics and protection from impacts and dust. 2.2.3 Product Timeline Table 1 below depicts the high-level view of the product development timeline for Daktari. It involves the development of two prototypes (gamma and alpha) that will be used to solve various anticipated issues. The final one i.e. beta will be the first production-level product. This thesis focuses on issues related largely to the alpha and beta stages. Table 1: Product Development timeline Purpose Production Quantities of: a) Cartridges a) Instruments Proof-ofconcept for investors 1 1 Run clinical trials and establish the reliability and repeatability of the test 1000s 10s First production product 350,000 (per year) 150 (approx.) 2.3 Problem Statement As Daktari moves from prototype to production i.e. alpha to beta, a number of issues will need to be addressed in order to ensure reliable, cost-effective production. Typically, the effect of manufacturing variation on the functionality of the proposed design must be understood. With this knowledge, a range of design parameters may be optimized leading to a more robust design. In some situations a conflict between design requirements and inherent manufacturing limitations may arise. In this case, suitable quality control strategies must be provided. All of these activities -both experimental and analytical - are commonly grouped together under the umbrella term 'product industrialization'. The absence of such a step often leads to bad designs which, when put into production, lead to poor quality, wasteful processes, high production cost and long lead times. Thus, it becomes important to understand that a good product is one that not only performs its intended function when 'perfect' but is also easy to manufacture, easy to assemble and, hence, economical to produce. The aim of this M.Eng. project is to analyze and propose solutions for issues related to the mechanical robustness of the Daktari system. Accordingly, a survey was first done to list all issues that will require attention en route to mass production. 2.3.1 Survey of Manufacturing Issues The Daktari POC system uses a number of components that have not been used extensively previously and so several areas require further study: a. The Electrode Foil Manufacturing process The electrode foils on the cartridge are critical to the sensing process of the LOC system. Preliminary studies indicate that repeatability of the impedance measurement can be strongly related to the exact electrode configuration and manufacturing defects. Further study is required to understand this phenomenon and ensure repeatability. b. Functionalization process The functionalization process refers to the process of depositing antibody solution on the electrode foil (see Section 3.4.2 for detailed description) . This process is again critical to the success of the assay because improperly coated antibody can result in several problems: * improper bonding between the electrode foil and the backbone leading to leaks * variability in the number of CD4+ cells captured by the system leading to low repeatability in results c. Injection Molding of the Backbone The injection molding process of microfluidic devices often requires very large normalization times - the time required for the plastic to reach its final dimensions - often in the order of days. Any downstream processing that affects normalization process or requires the dimensions that are affected by it cannot be performed until the normalization is complete. This results in long lead times which could be a potential issue in manufacturing. (resulting in large inventories and associated storage costs) Although, this issue could have a large impact on the manufacturing cost of the final product, currently it does not seem to be critical to the product functionality. d. Robustness of Instrument- Cartridge Interactions i) Blister - Actuator Interaction The interaction between the blister and the actuator produces the required flow in the final device. Precise movement of a precision-machined actuator ensures controlled flow. However, the amount of control obtained based on current design specifications and manufacturing variation is unknown and should be studied to confirm flow rate tolerances. ii) Valve - Solenoid Interaction The valve and the solenoid together ensure control over the direction of fluid flow within the cartridge. This is important as leaking valves (valves that are open when they should be closed) and 'sticky' valves (valves that are closed when they should be open) can seriously affect assay performance. A tolerance analysis of this step should be performed to determine the effect of misalignment. iii) Vent hole - Puncture Pin Alignment The vent hole is a weak portion of the backbone that is designed to be opened by a pin in the instrument. This is necessary to open the system to the environment allowing fluids to be run through without causing air compression. Failure to open the vent hole can be critical to product success as it may become nearly impossible for the actuators to pump fluid through the system. iv) Electrode Gold Pads - Electric Connector Interface The electrical connection between the electrode foil and the instrument is critical to the reading the results of the assay. A failed connection will prevent impedance measurements from being taken. However, this connection also has much tolerance built into it and it is not believed that it will cause much problem. The above cartridge-instrument interaction issues are summarized in Table 2. Table 2: List of Potential Cartridge-Instrument Interaction Issues Interface Tolerable Variation Approximate Manufacturing Variation Blister - Actuator Unknown Several mm Valve - Solenoid Unknown Several mm Vent Hole - Puncture Pin 0.5 mm (radial) Several mm Electrode Contact Pad - +/- 1 mm Several mm Instrument Electrical Connector 2.3.2 Selection of Subset of Problems Table 3 displays a list of the above issues along with a note about their importance (for product function) and likelihood of occurrence. From this list, issues related mainly to the mechanical robustness of the device were selected. These were all the issues under sub-section 'd'. Table 3: List of Potential Issues during Ramp-up Importance for Product Function Likelihood of Occurrence a. Electrode Foil Manufacturing High High b. Functionalization High High c. Injection Molding of backbone Low Low i) blister-actuator Medium Medium ii) valve-solenoid Medium Medium High Low Issue d. Robustness of Interactions iii) vent hole -puncture pin iv) electrode contactpatch - instrument contactpins High Low It was proposed that a detailed tolerance analysis of the entire system be performed to determine the overall stack-up of manufacturing variation on the various interfaces. In addition, the influence of this variation on the performance of the interfaces would be studied to determine their effect. In particular, greater focus was to be placed on the blister-actuator interface. A numerical model relating blister performance to the dimensional variations was developed in this thesis and this model was used in conjunction with the tolerance analysis of the entire system developed by Linares . Linares  also performed the tolerance analysis for the remaining interfaces. This study would lead to a better understanding of the interfaces and permit the suggestion of suitable solutions for optimizing the designs. 3. Background Research Initial research focused on understanding the applications of microfluidics (Section 3.1) and, thereafter, on studying the state-of-the-art in microfluidic device architecture (Section 3.2), components (Section 3.3) and manufacturing processes (Section 3.3). This was done in order to fully understand the interactions between this study on robustness and the design and manufacturing of Daktari's microfluidic device. After this preliminary research, the tools and techniques used in this thesis were explored. This includes the technique of Monte Carlo simulations used for variation analysis and the use of machine vision for metrology. 3.1 Lab-on-a-Chip Applications An extensive survey was done by Korb  on the various potential applications of microfluidics. It quickly became apparent that most applications were in biochemistry and related fields. The applications can be grouped as: assays, drug discovery, genomics and proteonomics, cytology and biotechnology, drug delivery and surface patterning. The CD4+ sensing system falls into the cytology and assay category. In cytology, microfluidics is used to sort cells and select ones of interest. In assays, the presence or concentration of a substance is determined. For example, the purpose of Daktari's microfluidic device is to perform an assay on a fixed volume of blood and determine the concentration of CD4+ cells. 3.1.1 CD4+ Cell counting through Cell Lysate Impedance Spectroscopy This is the technique used by Daktari in its CD4+ Cell Counter System. In this method (see Figure 4), cells are first immobilized by the use of cell affinity chromatography . Then, the cells are lysed, and a drop in impedance is produced due to the different electrical properties of the cellular contents. The change in impedance is measured using the electrode foil. It is converted to a change in conductance. The magnitude of this change increases linearly with the number of immobilized cells . These methods showed a close correlation between the CD4 cell counts when comparing the microfluidic device and flow cytometry . ya F1'i Y + CC Y Antibody Red Blood Cell CD4+ Cel Figure 4. Assay Process Diagram. (A) Blood is ran through the assay chamber and CD4+ cells are captured. (B) Red blood cells are washed. (C) CD4+ cells are lysed and difference on impedance is measured. 3.2 Microfluidic Device Architecture The architecture of microfluidic devices is similar to that of Printed Circuit Boards (PCB). These devices are formed by different layers each performing a specific function. Figure 4 shows a typical arrangement consisting of one or more central layers (backbone) where open micro channels are formed (usually with 2.5D features) and one or two external layers to close the channels. Additionally external connections, valves, pumps and sensors are often added. ............ . . .. .. ......... ....... Additional Component Central Layer External Layer Figure 5. Microfluidic Device Architecture 3.2.1 Central Layer This layer is often the functional part. All the features, such as microchannels, mixers, reservoirs and other components are formed in this layer. Its complexity varies from a simple piping system to a complex set of mechanisms. 3.2.2 External Layer Its purpose is usually to close the channels and features on the central layer. Additionally functionality is often added to this layer using PCBs, sensors and outputs from other systems. 3.2.3 Additional Components In addition to the above, microfluidic devices typically contain other components that perform external flow control, internal flow control and sensing. ExternalFlow Controlmechanisms are added to provide an external means of delivering reagents into the Lab-on-a-Chip (LOC). The work involved in pumping arises from non-microfluidic interactions. Some examples are external reservoirs (where the column of liquids drives the fluids), actuators ( which push liquid-filled blisters), external valve systems and syringe pumps. InternalFlow Controlcomponents fulfill the same purpose as those placed externally. However, flow arises due to interactions with microfluidic features. Examples include peristaltic or pressure driven pumps, valves, mixers, separators, reactors, etc. Sensing components are often put into place on the external layers of the device. Commonly these are used to measure pressure, temperature or electrical properties (impedance, capacitance, etc.). Additionally, much attention has been focused on optical sensing systems. 3.3 Lab On a Chip Technologies For a microfluidic device to perform any particular function, whether that be an immunoassay or genome sequencing, several proven technologies are currently available to the engineer or researcher that can be used as tools or components to achieve their goals. Among them we have microfilters, microneedles, micromixers, microreactors, microdispensers and microseparators. Nguyen  provides a very good description of these technologies and its applications in life sciences. Microfilters are used for either filtration or collection depending on whether the substance of interest is a liquid that is contaminated by particles or the particles themselves. Both of these functions are achieved through the use of membranes or gaps built into the microfluidic piping system. Microneedles are used for drug delivery, cell manipulation and interconnection between micro and macrofluidic systems. Micromixers combine two or more reagents to make a reaction possible. The advantage of this technique lies in the efficient use of reagents. The key difference between macro and micro scale is the mechanism of mixing - macroscale mixing uses turbulence whereas microscale mixing depends on much slower diffusion processes on account of the low Reynolds' number in microfluidic channels. Microreactorsaccomplish the same tasks as their macroscopic counter parts, but with several advantages. The reactions are easier to control, reducing the danger of an explosion or fire. Additionally if dangerous substances are released, they are easily contained. They offer a cost advantage both in manufacturing and in their operation. They are easy to scale from laboratory to commercial applications. Finally they offer small thermal inertia, high gradients, uniform temperature, short residence time and high surface to volume ratio. Microdispensersprovide precise reagent delivery. Examples include microscopic injectors, pipettes or dosing systems. Microseparatorsprovide substance or particle separation. Microfilters can be considered as a category of microseparator that discriminate by particle size and geometry. Microseparators take advantage of difference on weight, electrical, magnetic or thermal properties. 3.4 Lab-on-a-Chip Manufacturing Processes Each of the parts of the microfluidic device architecture can be produced using a multitude of processes. In the literature, most attention has been given to the central layer manufacturing process because all the functional features are often built into this layer. Additionally, and equally important, are the processes for assembly of the several layers. These processes involve alignment, placement and binding. The materials in which the devices are made greatly influence the process in which it is constructed. The most common material used for LOCs are glass, polydimethylsiloxane (PDMS) and polymethylmethacrylate (PMMA). Glass and PDMS are commonly used in laboratories since fast and precise prototyping techniques exist that allow researchers to iterate rapidly between different designs. Many of the techniques used in the semiconductor industry are borrowed to achieve nano-scale features. Also, many components of LOCs (e.g.: some kinds of valves and pumps) are only achievable on PDMS since they take advantage of the material's elastic properties. PMMA presents bigger difficulties for prototyping. Tooling fabrication is required, thus increasing the turnaround time. Nevertheless, since the 2000s there has been a growing consensus on the use of polymeric materials for microfluidic commercial applications . Attia et al  performed a useful comparison between the different materials used on microfluidic devices. 3.4.1 Part Manufacturing Korb  also performed a detailed survey on the manufacturing processes currently in use and their advantages and drawbacks. For central layer manufacturing, two methods were considered for this work due to their high throughput - injection molding and micro embossing. For external layer manufacturing, typical processes include rolling, molding and extrusion of thermoplastics. Injection molding is a well-known manufacturing method that can be used to produce complex forms out of thermoplastics. There have been successful attempts to replicate square features with a width of 310 nm and depth of 220 nm . It is highly suited to large scale manufacturing due to its dimensional control, short cycle times and high throughput. The challenge of bringing this technique to the micro scale lies in tooling construction. To the author's knowledge, injection molded microfluidic parts have been limited to having 2.5 D features. Micro-embossing involves raising the temperature of a blank polymer piece beyond its crystallization point followed by pressure to transfer features from a tool to the blank. The process has been shown to allow nanometer scale features  with high accuracy and precision. Its primary advantage over injection molding is in lower tooling costs and potentially lower cycle time. However, the process is limited in the complexity of features that can be created. Also, when an embossed part is subjected to high temperatures in subsequent operations (such as thermal bonding) the surface stresses tend to relax leading to a softening of the features. Overall, the use of injection molding appears to be preferred to the greater flexibility and the fewer manufacturing steps. 3.4.2 Functionalization Process Functionalization refers to the deposition of antibody on one or more parts of the product. The process is not well-understood and successful applications tend to be based on empirical experience rather than an actual understanding of the phenomena that take place. Antibodies can be deposited onto solid phases by three different mechanisms : 1. Adsorption to hydrophobic surfaces 2. Covalent attachment to activated surface groups 3. Non-covalent, electrostatic and hydrophilic bonds by either of the before mentioned. It is a requirement for Daktari that the antibody be deposited in a specific area of the electrode foil. Currently, this is accomplished using a masking process. The mask is placed on the electrode and the open cavities in the mask are filled with antibody solution. After a certain amount of time (termed 'incubation' time), the antibody solution is removed and the foil is allowed to dry. This multi-step process is illustrated in figure 6. Often, the antibody layer must be stabilized using a sugar layer as well. ...... ... ....... 1. Electrode Foil is covered with a mask 2. Antibody solution is added 3. Antibody is attached to Foil substrate 4. Excess solution is removed 5. Sugar solution is added 6. Sugar solution is dried 7. Mask is then removed -Foi Mask Figure 6. Functionalization Process 3.5 Monte Carlo Analysis Monte Carlo simulations are a technique for determining the parameters of an actual population by taking random samples from a pseudo-population and then calculating statistics using that sample. It involves the two steps of actually creating a pseudo-population of interest and then sampling from this population so that various statistics may be calculated. The Monte Carlo technique is widely used in situations where the behavior of the statistic does not lend itself to straightforward analytical evaluation. In such cases, Monte Carlo analysis can simulate the required sampling distributions allowing inferences to be made. The basic steps involved in Monte Carlo analysis are as follows : 1. The creation of a pseudo-population (usually a mathematical algorithm or model) 2. Sampling from the pseudo-population in a manner consistent with the real process 3. Calculate the value of the statistic from the sample 4. Repeat 1, 2 and 3 for 'n' trials 5. Plot a frequency distribution of the statistic which can then be used for inference This technique has been used in several studies [22-25] regarding manufacturability and tolerance analysis because of the complex nature of the interactions. In this thesis, Monte Carlo simulations are used to determine the effect of random variations in dimensions on the flow rate of reagents delivered by the blister packs. 3.6 Machine Vision for Metrology 3.6.1 Choice of Metrology System In order to accurately and repeatably measure the blister dimensions, two options were considered: using a coordinate measuring machine (CMM) and using an optical setup (a machine vision system). The CMM had the advantage of being more accurate (order of microns) and not requiring calibration. However, as the blister was a non-rigid part, and the CMM used a force sensor attached to a probe to measure the dimensions, its effect on the blister dimensions was unknown. As a result, the optical system was chosen. The optical system also had the advantage of having a much lower measurement time per blister. 5.1.2 Components of a Machine Vision System A machine vision system for measurement consists of two separate, yet inter-related, systems: the hardware and the software. The hardware involves all the parts of the machine vision chain that are responsible for producing a high-contrast, high-resolution image of the object that is being measured .The software is responsible for analyzing the image and obtaining the required dimensional information. HARDWARE SOFTWARE Data Acquisition Lighting Camera Computer Oiage Anasnis Output Mecasurements ObNect to be Measured Figure 7: Schematic of Metrology System using Machine Vision The hardware includes fixtures for repeatably mounting the objects to be measured, appropriate lighting that provides a suitable amount of contrast and a camera of sufficient resolution to capture the image. The camera is essentially a combination of the image sensor and the optical elements (lenses, mirrors, etc.) necessary to project the required part of the image onto it. The software includes the interface between the data acquisition system and the camera, the imageprocessing algorithms that enable suitable extraction of information and the final image analysis algorithms for measuring dimensions. 4. Development of the Blister Performance Model The blister performance model described in this section is used to relate the dimensions of the blister to its performance as measured by certain metrics. The process of creating the model began with an attempt at an analytical expression to model the working of the blister. However, the complex nature of the actuator and blister interaction quickly complicated the analysis and was abandoned in favor of a more practical approach using numerical integration. 4.1 Working of the Daktari Blister Pack Figure 8: Photograph of a Blister Pack (above) and Actuators (below) The blister pack (Figure 8) is that part of the cartridge which is used to store the liquid reagents used in the assay. It is similar to a blister pack that is used to store pharmaceutical tablets/capsules and is manufactured using a similar cold forming process. Figure 9 depicts the blister pack in action. An actuator is allowed to push down the blister until the seal breaks. Further movement causes the liquid reagent to flow into the micro-channels on the backbone. Controlled actuation should enable a constant flow rate through the micro-channels. The constancy of thisflow rate is the most important performance measure of the blister pack. I V I ACrUATOR BLISTER 1.The actuator begins 2. The actuator moves down towards the bhster with reagent at the 'home' position 3. The actuator comes into contact with the blister (shown in blue) i t 5. The actuator 6. As the actuator continues to move moves down, it slows downwards forcing reagent out of the blister down maintaining the same flow rate 4 4. The actuator moves down approximately 2.5mm to break the blister seal (in red) 7.The actuator stops moving when 184) micro liters of reagent flow out Figure 9: Blister actuation process For Daktari's current application, a tolerance of +/- 20% on the flow rate is believed to be acceptable. For example, for a nominal flow of 20 pL/min, an actual flow rate that ranges between 16 and 24pL/min is satisfactory. In order to maintain this constant flow rate, the speed of the actuator is decreased as it moves down, in order to compensate for the increase in crosssectional area (Since, volume flow rate = linear velocity * cross-sectional area). Very precise actuation is necessary to maintain a smooth flow and, to the knowledge of the author, such a procedure has not been studied till date. 4.2 Requirements of the Blister Numerical Model The blister model should be capable of sufficiently re-creating the working scenario described in the previous section. Taking the actual dimensions of the blister and actuator as input, the model should be able to give the flow rate of the fluid as a function of the position of the actuator (See Figure 10). Depth Cru A = B = Total Volumc Expcllcd A at crush dcpth a6 .E E Crush Depth (mm) Figure 10: Required Outputs from a Mathematical Model of the Blister In addition, for purposes of studying the effect of manufacturing variation on the blister performance, the model should include the continuously varying speed of the actuator as it plunges into the blister. y Y6 ;44Y!,! " 4.2.1 Blister Outputs and Noise Factors The total volume of fluid required from each blister is 180pL. As the actuator tip plunges into the blister, this 180pL of reagent should be expelled at a constant flow. Thus, the function of the blister can be captured by the following two metrics: 1. Average flow rate: This captures the mean value of the flow rate. 2. Maximum of flow rate: This gives the upper bound of the flow rate due to nonconstancy of flow 3. Minimum of flow rate: This gives the lower bound of the flow rate due to nonconstancy of flow The above metrics are depicted in Figure 10. The various variables, which affect the flow rate of reagents in the channels, are as follows: 1. Base Radius 2. Spherical Radius 3. Blister height 4. Lateral Misalignment 5. Angular Misalignment 6. Actuator Radius 7. Actuator's Starting Height (i.e. initial position of the actuator above the blister) Figure 11 depicts variable 1 -3 of the above list. These are related to the blister pack production process while 5-7 are related to the instrument manufacture. Variable 4 is related to both. For the purposes of this analysis, variables 1 through 4 are taken as noise factors. 0 Blister Radius Figure 11: Particular blister dimensions that affect the flow rate --- 4.3 The Numerical Model The numerical model is based on the use of numerical integrationto calculate the volume of the blister at each step of the actuator. The motion of the actuator was discretized i.e. each step movement of the actuator was associated with a particular amount of time. The blister itself was divided into a set of discrete volume elements and its volume calculated at each step to determine the volume of fluid expelled. Figure 12 is the flowchart of operations that must be performed by the blister model. Each step is further discussed in the following sections. START Input Blister and Actuator Dimensions and Positions Calculate TOTAL VOLUME EXPELLED up to the current step Input Speed of Travel LSet Calculating VOLUME EXPELLED IN CURRENT STEP by subtracting the total volume expelled up to the previous step Step Size =0.1mm Determine all ACTUATOR POSITIONS using transformation matrices Have all steps been completed? Calculate INSTA NTANEOUS FLOW RATE b y dividing the volume expelle d by the time taken by the actuator to complete the step No A Yes Calculate the AVERAGE, MAXIMUM AND MINIMUM FLOW RATE Output the above flow rates STOP Figure 12: Flowchart of Blister Numerical Model 4.3.1 Input Blister Dimensions and Speed of Travel The blister and actuator dimensions listed in Section 4.2.1 are provided to the mathematical model in this step in order to re-create the geometric shape of the blister. The required flow rate (5iL/min or 20RL/min based on the application) is also provided at this stage so that the appropriate actuator movement curve is chosen. 4.3.2 Step Size The 'step size' refers to the distance the actuator 'moves' during each step of the blister model. By comparing Figure 13 and Figure 14, the effect of having large step sizes can be seenvariations in flow rate which are actually present can get smoothened out by averaging over a larger range of movement. Thus, smaller values of the step size lead to better approximations of the actuator movement at the cost of increased computation time. An optimal step value was chosen based on trial and error. Volume Expelled per unit time (uL min) Au 13 Figure 13: Effect of small step size on flow rate measurement Volume Expeled per unit time Af- 06 " Figure 14: Effect of large step size on flow rate measurement 4.3.3 Determine ActuatorPositions accordingto Step Size and Misalignment X X YA The actuator position is based on an ideal blister and so crush depth is always measured downwards from the actuator axis (in blue) in order to account for the errors, the actuator is given an initial position and angular misalignment from the vertical Figure 15: Introducing the effect of actuator misalignment and step size In the physical system, the absence of a feedback loop to control actuator position could lead to problems due to misalignments (shown in Figure 15). In order to study the effect of the misalignments, they were included as inputs to the blister model at this stage. Both lateral and angular misalignments were included using a set of transformation matrices (see Figure 16) to set the initial position and angular misalignment separately. These matrices convert the unidirectional movement of the actuator into a 2D movement in real space. B B *Tab UL.~ * X Tac Tac transforms the coordinate system C to the blister coordinate system A The product of Tac and Tcb gives Tab which transforms the actuator coordinate system (B) to the blister coordinate system (A) '- Teb Tcb transforms the coordinate system from the actuator (B in blue) to the coordinate system C Figure 16: Usage of transformation matrices to convert actuator coordinates to the blister coordinate system Here, Teb is the angular transformation matrix that re-orients the actuator coordinate system in the same direction as the blister coordinate system. Then, it gets re-positioned to the blister coordinate system by means of Tac, a positional transformation matrix. 1I 1mO Tc = 0 0 HO 1] T = cos 6 sin 6 -sin 6 cos 6 0 0 .. ... ..... Thus, any coordinate in the actuator coordinate system (0,Actuator Depth), can be converted to the blister coordinate system as follows: Y = TIC * T, * (Actuator Depth The (x,y) coordinates of the actuator center obtained using the above expression can be used to calculate the total volume of fluid expelled. 4.3.4 Calculate the total volume offluid expelled up to the current step ACTUATOR BLIS7ERdV =Common Area x dY Figure 17: Discrete Volume Calculation In this step, the volume of fluid expelled from the blister is calculated using the process of numerical integration. As shown in Figure 17, the common volume between the blister and the actuator is divided into discrete volumes and summed together to obtain the total volume. For each volume slice, the common area between the two curves is determined and then multiplied by the height of the volume slice. The sequence of steps involved in the common volume calculation is shown in Figure 18. Input Blister and Actuator DIMENSIONS and POSITIONS Calculate BLISTER RADIUS RI at the height of the cross-section Calculate ACTUATOR RADIUS R2 at the height of the cross-section Calculate COMMON AREA using only smaller circle and multiply by dY to get dV Calculate COMMON AREA using circlecircle intersection and multiply by dY to get dV Figure 18: Flowchart for Blister Volume Calculation Provided the blister and actuator are present at a particular slice, the common area between the two can be easily calculated by the procedure depicted below in Figure 19. At a particular height, the intersection of blister and actuator produces two intersecting circles The blister circle (in green) and the actuator circle (in blue) intersect and the common area represents the portion of fluid expelled The top view of the intersection shows the common area (in yellow) Figure 19: Intersection of Actuator and Blister at a particular height There are 3 possible scenarios when the blister and actuator circles intersect: 1. The circles do not intersect Figure 20(a) shows a situation wherein the two circles do not intersect. In this case, the common area is zero and this volume slice does not contribute to the volume calculation. iw - +m _ _ Ar _ NMMMMMhF_ _::_: RI +R2<d RI+R2>d>R2-RI d<R2-RI Non- Intersecting Circles (a) Intersecting Circles (b) Circle within another circle (c) Figure 20: Possible Scenarios when a Blister and Actuator Intersect 2. The two circles intersect Figure 20(b) shows a situation where the two circles intersect. The common area between the two circles is calculated using a circle-circle intersection formula  The horizontal distance between the points of intersection (P,Q) and the center of the smaller circle 01 is denoted by 'x'. d2 - R2 2 2d + R2 Key: 00, RI - Center-Center Distance, 'd' - Radius of the Blister Circle R2 - Radius of the Actuator Circle Figure 21: Calculating the area of a Circle-Circle Intersection The area of intersection between the two circles is obtained as the sum of the areas of the asymmetric lenses (shown in blue and yellow in Figure 21) formed by the two circles. Common Area = A1 + A2 where d 2 A2 =dR2 cos-1( - x) R2 x - (d-x) R2 2 - (d - x)2 3. One circle is entirely within another Figure 20(c) depicts a situation wherein the smaller circle lies entirely within the larger. In this case, the area of the smaller circle gives the common area between the actuator and blister at that point. This is calculated simply as: Common Area= ; R 4.3.5 Volume expelled in the current step The volume calculated from the previous steps represents the crushed volume of the blister i.e. the volume of fluid that has been expelled up to that point. From this step, the volume expelled during any step can also be found simply by subtracting the total volume expelled up to the current step (Figure 22). Volume Expelled in Current Step = Total volume expelled up to current step - Total Volume Expelled at Current Actuator Step Total volume expelled up to previous step Total Volume Expelled at Previous Actuator Step Volume Expelled during the Current Step Figure 22: Volume Expelled in the Current Step 4.3.6 Getting the 'instantaneous'flow rate Once the volume expelled in a particular step is found, it is divided by the time taken by the actuator to complete that step. This time depends on the position of the actuator and is calculated so that an ideal actuator in contact with an ideal blister will produce constant flow. ACTUATOR Total Volume Expelled (uL) Crush Depth (mm) BL ISTER dH Total Volume Expelled = f (Crush Depth) Crush Depth 0 (mm) dH - step size dV - volume expelled in current step Figure 23: Total Volume Expelled vs. Crush Depth Figure 23 shows how the time taken for a given step is calculated from the ideal blister-actuator interaction. For a particular step 'dH', the corresponding volume expelled is given by 'dV'. This volume must be expelled in a certain period of time 'dt' such that the flow rate is a constant. -, V dt where V is the constant flow rate required dt = dV . , for that particular step dH corresponding to a particular crush depth 'H' This value dt is determined for the entire range of movement of the actuator and plotted (Figure 24). Using regression. -t dt=f (H) Crush Depth (H) Figure 24: Time taken for a Step versus the Crush Depth A regression equation of dt vs crush depth is obtained from this graph and using this the time taken for any step can be calculated. Once the time dt is known, the instantaneous flow rate is calculated as: InstantaneousFlow Rate = Volume Expelled in CurrentStep (dV) Time takenfor Step (dt) 4.3.7 Calculate the average, maximum and minimumflow rates From the set of instantaneous flow rates obtained from the previous steps, the average as well as extreme flow rates are determined. These metrics (as discussed in Section 4.2.1) are taken to be representative of both the mean and non-constancy of the flow profile. These metrics once calculated are the required outputs of the blister numerical model. 4.4 Assumptions The following assumptions were made in the development of the blister flow rate model: 1. The blister collapses regularly with no crumpling This assumption greatly simplifies calculating the common volume between blister and actuator. In situations where crumpling of the blister begins to occur, the uncertain nature of crumpling leads to sudden increases in flow rate which are difficult to predict. It has been shown experimentally (by Daktari) that the current blister and actuator shape do not cause crumpling. However, this may not be the case for arbitrary shapes of blister and actuators. A great deal of optimization may be required to arrive at a suitable configuration. 2. There is no air in the blister Air present in the blister tends to act like a damped spring (due to its viscosity and compressibility). This could lead to large time constants before reaching stable flow. By assuming the absence of air, these dynamic effect are taken to be absent and the liquid responds instantaneously to changes in the velocity of the actuator. 3. Angular misalignments are measured about only a single axis. This blister model makes the assumption that angular misalignment and lateral misalignment occur in the same plane i.e. only 2d misalignments have been considered. While it is believed that this is representative of the real system, a more accurate model would include another angular misalignment in order to fully study the different modes of variation. 5. Validation of the Blister Model Validation of the blister performance model developed in the previous section was attempted using two setups: an optical setup for measuring the blister dimensions and an experimental setup for measuring the flow output from the blisters. This data was correlated and compared to the model predictions to determine the model accuracy. 5.1 Measurement Setup for Blister Dimensions 5.1.2 Selection of Hardware Fixturing of a cartridge (containing a blister pack, see Figure 2) was accomplished by fixing the plane of the cartridge on a base and thereafter locating using a 2-1 fixturing system. Figure 25 shows a schematic of the setup used. An interesting feature of this system is the ability to index the 'blister locator'. This allows both top and side views of each blister to be photographed without having to change the focus. Top View Side View LiZUZ1 Blister Locator Top View Side View I- 0 Figure 25: Schematic of Blister & Camera Setup The camera used for the photographs was a Fujifilm FinePix S700 with a Fujinon Zoom Lens (f= 6.3mm). Although the resolution of the camera was around 7.1 megapixels, the zoom lens allowed a large portion of the image to be occupied by the blister increasing the number of pixels devoted to it. Figure 26 is a photograph of the entire setup with camera and cartridge mounted. Top View of the Setup Thc blister locator is in position to measure the side profile of the blisters Front View of the Setup The blister iocator is in position to photograph the top view of the blisters Figure 26: Photograph of Blister Measurement Setup Calibration of the setup indicates that each pixel corresponds to a distance of 17.6 microns at the focal plane. Using this setup highly repeatable photographs could be taken. Figure 27 shows some examples. Side -vicw of the blister pack showing the height h)aand sphecmal raditus (r i Iop iew of the bhter pack shouing the base radius (R) Figure 27: Photographs of the Blister Using the Current Setup 5.1.3 Selection of Software Typically the software in a machine vision system, is expected to perform quick analysis of an image and output the required dimensional information in a manufacturing line. However, for the purposes of this project, time of measurement was not a significant constraint. Thus, instead of using image processing algorithms to extract features from a captured image, it was decided to manually count the number of pixels associated with each dimension. Although, this process is time consuming and should be avoided for higher number of tests and higher repeatability, it has been adopted in order to minimize the time required to setup the image processing algorithms. In order to determine the repeatability of manual measurements, a gauge repeatability and reproducibility (R&R) experiment was performed 22.214.171.124 Gauge Repeatability and Reproducibility The repeatability of image manipulation and the entire measurement setup were measured in two separate steps. For the repeatability of image manipulation, the same images of a blister were measured 5 times by 2 operators. The results are shown in Figure 28 and Figure 29. Blister Diameter from Top View 522 $21 520 519 518 517 Sib * 521 £ 5-20 4 ft-i $20 * 516 1 Operator No. Figure 28: Blister diameter (in no. of pixels) measured by 2 operators Blister Height from Side View 1713 172 170 169 168 161 166 ---*IM * 168 £170 --- 1698 *18 Operator No. Figure 29: Blister Height (in no. of pixels) measured by 2 operators This overall standard deviation between the two operators was found to be 1.44 pixels which translates into a repeatability coefficient of 4.3 pixels i.e. 75.68 pm. This was deemed acceptable for the experiment. 5.2 Measurement Setup for Blister Flow Blisters measured using the above setup were driven by an actuator in order to determine what the flow profiles coming from them were like. This data was then used to correlate the measured dimensions with the final flow rates. The setup used for this purpose was a test system used by Daktari Diagnostics for testing the working of their cartridges. This setup consisted of the actuators as well as the locating mechanism for the cartridges. A Sensirion SLG1430-480 flow sensor was used to measure the liquid flow rates from the blister. 5.3 Validation of Blister Numerical Model In order to validate the model, the dimensions of a set of 4 blisters were measured using the optical setup described in Section 5.1. Only the blister height and base radius were measured from the side and top views of the blister. These dimensions were used as input for the blister model, and the flow rates at specific crush depths was noted. These flow rates were then compared with those obtained from experiment. Figure 30 summarizes the results. . . M . .. ........................................ 24 22 14 - 12 3.5 3.7 3.9 4.1 4.3 4.5 4.9 4.7 5.1 5.3 Crush Depth (mm) 3A - Actual Flow 3B - Actual Flow 3C- Actual Flow -2B - Actual Flow 3A - Simulated Flow 3B3 - Simulated Flow "" 3C - Simulated Flow """ 2B - Simulated Flow Figure 30: Plot of Experimental and Predicted Flow Rates from 5 Blisters . The fit between the model and the experimental values is not satisfactory. There are several possible reasons for this: 1. Uncertainty in known dimensions The uncertainty in the known dimensions i.e. radius and height are high enough that it is possible that wrong input was being provided to the model. 2. Lack of information about other relevant dimensions of the system Data regarding the other inputs to the model was not available. In particular, the lateral misalignment and starting height of the actuator were not completely known. In the absence of this data the expected values for these dimensions were used in the blister model. However, flow rate variation caused by these dimensions has not been modeled. 3. Observed Lateral Misalignment Significant lateral misalignment was observed in the test setup at the time of flow measurement. However, in the absence of actual data this information could not be used. 4. Crumpling in the blister Some crumpling of the blister did occur during these trials, and this could have led to some of the differences between the model and reality. It is important that these effects are considered and the blister validation process repeated in order to verify the claims that follow. 6. Monte Carlo Analysis and Regression Models Despite the lack of a successful validation experiment (see Chapter 5), it was decided to use the model in a statistical simulation to determine what the behavior of the entire population of blister packs was likely to be (when manufacturing began and provided the model was correct). The method used is commonly known as the Monte Carlo technique. It involves the creation of a pseudo-population of blisters and actuators and then actually plunging these virtual actuators into the virtual blisters to produce virtual flow! A distribution of the flow parameters was built by running thousands of virtual experiments. 6.1 Manufacturing Variation Data The pseudo-population of blisters and actuators is built using data that informs us about their statistical distribution and parameters. Table 2 shows the manufacturing variation data for the various dimensions (see Section 4.2.1). Table 4: Variation Data for the Blister and Actuator Dimensions Standard Deviation Mean Value Distribution Dimension Base Radius Normal 7.99 mm 80 pLm Spherical Radius Normal 6.2 mm 80 tm Blister Height Normal 6.16 mm 35 tm Actuator Radius Normal 7.27 mm 33 pm Lateral Misalignment Normal 0 mm 142 jim Angular Misalignment Normal 0 degrees 0.5 degree Height of the Actuator Uniform 13.43 mm 486 pm This variation data has been obtained from a variety of sources. The base radius, spherical radius and blister height are obtained from a combination of manufacturer data and the validation data from Chapter 5. The actuator radius data was selected based on a standard deviation that would provide 99.7% of produced parts conforming to the DIN 2768-1 specification. The lateral misalignment, angular misalignment and height of the actuator are obtained from the analysis of the instrument by Linares Error! Reference source not found. 6.2 Monte Carlo Algorithm Following the generation of pseudo-population data, the algorithm consisted of using the blister numerical model developed in Chapter 4 for plunging the blisters and determining the flow rates from each setup. The algorithm is summarized below in Figure 31. START Generate a random blister and actuator using the population parameters for each Input NUMBER OF TRIALS, N Use the blister numerical model to calculate AVERAGE, MAXIMUM AND MINIMUM FLOW RATES Input Blister and Actuator Population Parameters Set Counter = I No Yes Counter <= N? Store the average, maximum and minimum flow rates in suitable vectors Counter = Counter + I Display a histogram of the stored flow rate data Determine appropriate statistics from the above simulations STOP Figure 31: Flowchart of the Monte Carlo Algorithm Several different experiments were run using the above Monte Carlo simulation and their results are summarized in Chapter 7. 7. Results and Discussion Using the Monte Carlo algorithm described in the previous chapter, a simulation of 10,000 blisters was conducted to study the effect of variation on the flow rate. Based on the requirements set forth by Daktari, the tolerance on the flow rate was set at +/- 20% of the mean flow rate. Any blister-actuator configurations that caused a flow rate outside this range were deemed 'out-ofspecification'. For example, for a nominal flow rate of 10ptL/min, the flow would be considered within spec if it were between 8piL/min and 12 pL/min. 7.1 Flow Rate Variation after considering only Blister Dimensional Variation To begin with, the robustness of the blister design with respect to an ideal actuator was studied. This involved allowing the 3 blister dimensions, namely, base radius, spherical radius and blister height to be randomly picked from suitable normal populations (see Table 4) while the actuator dimensions remained constant at their mean values. Figure 32 shows the distribution of flow rates for this situation. The three distributions capture the important elements of the overall flow rate profile shown in Figure 10. Table 5: Summary of Output Distribution Properties considering the variation in blister dimensions alone Average Flow Rate Maximum Flow Rate Minimum Flow Rate Mean 20.0041 20.0896 19.9166 Standard Deviation 0.2293 0.2404 0.2369 Skewness 0.0333 0.4955 -0.4477 Coefficient of Kurtosis 2.9592 3.3527 3.2543 Lilliefor's Test for > 0.5 < 0.001 < 0.001 Normality (p-value) 400 Frequency Dsrbution of Average FlowRat 300 100 r Figure 32: Distribution of Flow Rates considering only Blister Variation The distribution of average flow rate looks quite normal. The skewness, co-efficient of kurtosis and Lilliefor's test agree with this hypothesis (see Table 5). The distributions of maximum and minimum flow rates look like right- and left- skewed normal distributions. During the above simulation, no out-of spec parts were found. This seems to indicate that the interaction is robust with >99.99% conforming parts. This simulation is useful because it provides a lower bound on the capability of the blisteractuator manufacturing process. If the number of non-conforming parts was unacceptably high, it would have been impossible to improve manufacturing outcomes without adjusting the blister forming process itself. However, for current applications, the forming process provides a robust output and so it can be accepted. Using the above data, it is possible to show that the blisters can provide a satisfactory quality (assumed to be 3a i.e. 99.7% conforming parts) for a tolerance range as tight as ± 5%. However, as flow rate tolerance requirements get tighter beyond this number, further control of the blister manufacturing process is necessary. This may be in the form of greater process control, leading to less variation in blister dimensions, or it could be in the form of 100% inspection, which would provide a similar result but with greater yield loss. This data is summarized in Table 6. An interesting point concerns the process capability index Cp. For a centered process, CP =USL - LSL 6u where a is the estimate of the standard deviation for an approximately normal distribution. In order to satisfy that normality requirement, the average flow rate from the blister has been used to compute the C, values. It is interesting to note that nearly 6c capability is easily possible based on the average flow rate alone (C, = 1.45 for L 5%). However, in the current system, the maximum and minimum flow rates are assumed to play a critical role. If the tolerance requirement is placed only on average flow rate or some kind of external mechanism could be introduced into the microfluidic chip that would dampen out the fluctuations in fluid flow, it may be possible to get much higher capabilities using the existing process. Table 6: Increase in Percentage of Non-conforming Blisters with tighter tolerance ranges Required Flow Rate Tolerance i 20 % i 10 % ± 5% ± 2.5 % ±1 % Percentage of Non-Conforming Parts ~0% ~0% 0.14% 11.14% 58.59% Process Capability Index (Cp) 5.81 2.91 1.45 0.73 0.29 7.2 Flow Rate Variation after considering the variation in all dimensions The entire system including actuator variations may be simulated to obtain a better understanding of the system. Figure 33 shows the distribution of flow rates for this situation. Frequency Distbtlon of Averge Fow Rates (pirnin) Figure 33: Distribution of Flow Rates considering variation in all dimensions It can be seen that the addition of other sources of variation (due to the actuator and the assembly process) has dramatically increased the range of flow rates. The range of flow rates has increased to between 6 and 30 pL/min. The number of out-of-spec parts in this run of 10,000 blisters is 2438 i.e. 24.38% of the parts! This is certainly an unacceptable scenario. Table 7 shows a summary of the distributions' parameters. Table 7: Summary of Output Distribution Properties considering the variation in all relevant dimensions Average Flow Rate Maximum Flow Rate Minimum Flow Rate Mean 19.9837 20.7378 19.1504 Standard Deviation 2.2246 2.5407 2.6435 Skewness -0.0546 0.7708 -0.9171 Coefficient of Kurtosis 3.0195 3.3349 3.7907 >0.5 < 0.001 < 0.001 Lilliefor's Test for Normality (p-value) In this case, with no further intervention the tolerance limits on flow rates that are possible are: Table 8: Increase in Percentage of Non-conforming Blister-Actuator Systems with tighter tolerance ranges Required Flow Rate Tolerance ± 20 % ± 10 % ± 5% ± 2.5 % ±1 % Percentage of Non-Conforming Parts 25.15% 57.11% 77.91% 89.48% 97.86% Since, this kind of performance is not acceptable (even for the current tolerance of ± 20%), further study was done to determine the parameters that critically affect the flow rate. 7.3 Regression on the Average Flow Rate Using the data from the previous simulation, a regression equation for the average flow rate from a blister-actuator combination was extracted. A first-order regression with only main effects and no interaction effects has been found to be suitable to 0.6% accuracy. Table 9 shows the coefficients of the corresponding terms. Table 9: List of Coefficients for Linear Regression on Average Flow Rate Normalized Dimension Coefficients Percentage Contribution Base Radius, R 0.0228 1% Spherical Radius, r 0.1644 6% Blister Height, h 0.1494 5% Actuator Radius, Ra 0.1923 7% Lateral Misalignment, im -0.0004 0% Angular Misalignment, am 0.0006 0% Starting Height, Ho -2.1907 81% Table 9 shows the parameters that had a significant effect on the blister variation. They are highlighted: spherical radius (r), blister height (h), actuator radius (Ra) and starting height of the actuator (Ho). The final regression equation is for average flow rate is: Average Flow Rate = 19.9703+ 0.1643 r -Mp' + 0.1496 ( (Oh r 2 +0.1923 (Ra -1Ra \aRa / HO - pH aH ) 7.4 Regression on the Range of Flow Rates Here, the range of flow rates refers to the difference between the maximum and minimum flow rate for each blister-actuator combination. The coefficients for a first-order regression equation on the range using only main effects and no interaction effects is shown in Table 10. Table 10: List of Coefficients for Linear Regression on Range of Flow Rates Normalized Dimension Coefficients Percentage Contribution Base Radius, R Spherical Radius, r Blister Height, h Actuator Radius, Ra Lateral Misalignment, lm Angular Misalignment, am Starting Height, Ho 0.0114 0.0441 0.0061 0.0057 0.0113 0.0079 0.0822 7% 26% 4% 3% 7% 5% 49% This regression model indicates that the spherical radius and starting height have the largest effect on the range of flow rate. However, the other dimensions also have appreciable effects on the range of flow rates. 7.5 Effect of Variation in Individual Dimensions on Flow Rate In order to further understand the errors in a given blister-actuator configuration, the effect of each individual error has been studied. Each graph in this section has been plotted for the 2 sigma boundaries of the individual variables i.e. jt+2a and p-2a. The results are presented as follows: 7.5.2 Effect of Spherical Radius Variation Variations in spherical radius lead to the top sphere being either larger or smaller than ideal. We would expect this to affect the flow rate near the beginning of the actuator's movement. This is indeed the case as shown in Figure 34 and Figure 35. The flow rate starts off away from the ideal rate of 20 pL/min and slowly converges towards the final rate. The two step changes appear to take place near where the spherical and conical parts of the blister join together. The average flow rate is approximately 2% different from the ideal flow rate and the range is small. 21 - S20.5 ~20 a 19.5 0 19.I 2.5 3 3.5 4 4.5 5 Crush Depth (mm) Figure 34: Flow Rate Profile with a Smaller Spherical Radius 5.5 21 T!20.5 a 20 49 M 19.5 a: 0 19 2.5 3 3.5 4 4.5 Crush Dpth (mm) 5 5.5 Figure 35: Flow Rate Profile with a Larger Spherical Radius 7.5.3 Effect of Blister Height Variation When blister height changes without alteration in the other dimensions, the shape of the blister changes such that the cross-sectional areas are either always higher or lower than ideal. This results in a flow that starts off lower or higher and slowly but smoothly converges towards the ideal rate (Figure 36 and Figure 37). The average flow rate is approximately 1% different from ideal. The range of flow rates is also small. 21 20.5 _4 20 S19.5 0 2.5 3 3.5 4 4.5 5 Crush Dp (mm) Figure 36: Flow Rate Profile with a Smaller Blister Height 5.5 Figure 37: Flow Rate Profile with a Larger Blister Height 7.5.4 Effect of Actuator Radius Variation Changes in the radius of the actuator result in changes with the point at which flow begins. Thus, a larger actuator starts flow earlier than expected and leads to higher flows at every depth due to the larger intersecting area. Thus, overall shifts in the mean flow from ideal seem to indicate actuator radius variation. This is shown in Figure 38 and Figure 39. The mean flow rate is approximately 2% different from the ideal flow rate and the range of flow rates is still small. I I I I I Figure 38: Flow Rate Profile with a Smaller Actuator Radius 20.5 20 19.5 0 2.5 3 4 3.5 4.5 5 5.5 Crush Dept (MM) Figure 39:Flow Rate Profile with a Larger Actuator Radius 7.5.5 Effect of Starting Height Variation Variations in starting height affect the speed at which the actuator moves through the blister. If the actuator starts at a higher than ideal location, then the speed profile will be slower than the ideal flow rate and vice versa if the actuator starts off at a lower position. This change is however quite dramatic. Figure 40 and Figure 41 show large changes in flow rate. The mean flow rate is approximately 20% different from the ideal value and the range of flows is also quite large (approximately 4 pL/min). ~25 1-6 a0 2.5 3 3.5 4 4.5 5 Crush Dep (MM) Figure 40: Flow Rate Profile with Lower Actuator Starting Height 5. . Figure 41: Flow Rate Profile with Higher Actuator Starting Height 7.6 Flow Rate Variation after consideringthe effect of shimming Based on the analysis in section 7.5, it can be seen that the most critical parameter affecting flow rate is the starting height of the actuator tip. Having understood this, Daktari's current design already incorporates the reduction of instrument variation in one important way - shimming the actuator in the direction of motion to correct for assembly variations. This reduces the variance in the dimensions to less than the size of the shim. Interestingly, the use of shims also changes the distribution of the actuator height variation from normal to uniform[ 14]. Using these changed numbers in the simulation the results shown in Figure 42 and Table 11 are obtained: Table 11: Summary of Output Distribution properties considering the variation in al relevant dimensions Minimum Flow Rate Maximum Flow Rate Average Flow Rate Mean 19.9996 20.1199 19.8155 Standard Deviation 0.2981 0.3226 0.3150 Skewness 0.0159 0.4457 -0.3699 Coefficient of Kurtosis 3.0098 3.1973 3.1528 > 0.5 < 0.001 < 0.001 Lilliefor's Test for Normality (p-value) . . ..... Figure 42: Distribution of Flow Rates considering the effect of shimming The distributions of flow rate show a distinct decrease in variation due to the increased control over the initial height of the actuator. During this simulation, no out-of-spec parts were produced for the run of 5000. The number of non-conforming parts for this situation is: Table 12: Percentage of Non-conforming Blister-Actuator Systems after shimming Required Flow Rate Tolerance ± 20 % ± 10 % ± 5% ± 2.5 % ±1 % Percentage of Non-Conforming Parts < 2.12 % < 2.12% 2.12% 28.46% 81.66% 8. Conclusions and Recommendations In conclusion, based on the current requirements, the blister design has been found to be robust to manufacturing variations. The blister pack should be capable of providing a controlled flow rate of L 20% satisfactorily with less than 2.12% of the blister-actuator pairs failing to meet this spec (although more number of simulations may indicate that this number is lower). However, future diagnostic products may require tighter tolerances on the flow rate. In that situation, it may become necessary for Daktari to better control the following critical parameters: * Actuator radius and starting height: Further control over these two parameters would enable flow rate tolerances as small as +/- 5% to be achieved - Blister height and spherical radius: As tolerance requirements become approach +/- 2.5%, the number of non-conforming parts quickly begins to escalate (see Table 6). At this point, it will become necessary to reduce manufacturing variation in the blister height and spherical radius. This could involve greater investments in the blister manufacturing process which would improve the process capability index values (Cp) listed in Table 6. 76 9. Future Work 9.1 Blister Model Validation Although a Monte Carlo simulation was run and the results and conclusions of that experiment were presented, it is still necessary to validate the underlying model. A more accurate and precise measuring system should be designed and used with a more repeatable flow sensor. Such a validation study could be an extension of the work done in Chapter 5. 9.2 Study of Dimensional Variation in Blisters The validity of the Monte Carlo results depends on the assumption of independence and normality of the blisters' dimensions. This hypothesis must be verified to be true by measuring and analyzing the dimensions of a large sample (usually n > 30) of blisters. Since, a common forming tool is used for making the blisters, it is possible that there is some correlation between the dimensions and the effect of this should be included in future analysis if that were the case. Also, the normality assumption, while true for most cases, must be verified. 9.3 Increasing the Number of Runs in the Monte Carlo Simulation Due to the time-consuming nature of the Monte Carlo analysis a limited run of 10,000 blisters was studied in order to estimate the parameters of the population of blister-atuators. However, a better understanding of the population can be had by running the simulations a greater number of times. 9.4 Electrode Foils - Configuration Study and Process Analysis Preliminary studies indicate that the presence of manufacturing defects on the electrodes does affect repeatability and, hence, further study and optimization of the process is deemed necessary. One possible strategy involves determining the effect of the defects and using this information to optimize the final process. Another approach (that is more aligned with the process of development) is to directly begin searching for alternate manufacturing processes that could eliminate most, if not all, defects. If such a process were found, then experiments should be performed to determine if repeatability is achieved. For example, it is believed that the use of an 'adhesion' layer could dramatically reduce the number of manufacturing defects. Another useful exercise would be the design optimization of the electrode configuration (size, shape and spacing of electrodes). It may be possible to optimize these parameters to produce a more economical (less metallic material) and more robust design. 9.5 Functionalization Process Optimization As seen in Section 2.3.1 .b, the process of depositing antibody on the electrode foil requires attention to correct for improper coating. Various manufacturing processes (such as micro-contact printing, masking, spray deposition) need to be explored to determine their capabilities. Finally, a process can be selected and its parameters optimized for Daktari's specific requirements. As mass production begins, the time taken for the functionalization process needs to be optimized. It is currently perceived that this step will have the second longest processing time after the injection molding process. A considerable reduction in processing time would lower inventory levels at this step. 9.6 Injection Molding of the Backbone - Normalization Time Optimization The actual time required for normalization of the injection molded parts must be studied experimentally and this information needs to be used to optimize the process. Also, the tolerances on the critical dimensions of the cartridge could be studied to determine if they can be widened without sacrificing product performance- this would directly lead to a decrease in normalization time. 9.7 Valve-Solenoid Interaction Robustness Study Currently, the actual sensitivity of the valve-solenoids interaction to misalignments is not precisely known. Experiments to determine this sensitivity should be performed. This information can then be combined with the tolerance analysis of the valve-solenoid DFC performed by Linares to confirm whether the interaction is robust or further design changes are required. 9.8 Effect of Imprecise Actuator Movement The actuators that are used currently to push the blisters are quite precise. The effect of less precise actuators (larger step size) on the flow rate performance of the blisters could be studied. If satisfactory, these larger-step-size actuators would lead to significant decreases in product cost. Unfortunately, large actuator steps tend to lead to high changes in flow rate (although the average flow rate would be the same). However it may be possible to introduce a certain amount of capacitance into the system (either in the actuator or in the blister) to smoothen the flow. 80 References  Korb, S.N., 2004, "Towards the Manufacturing of Microfluidic Devices : Fluid Flow in Multilayer Devices as a Test Case", S.M. Thesis, Massachusetts Institute of Technology, Cambridge MA.  Linder, V., 2007 "Microfluidics at the crossroads of diagnostics", The Analyst, 2007, 132, pp. 1186-1192.  Janossy, G., Mandy, F., O'Gorman, M., 2008, "Diagnostics in the Shadow of HIV Epidemics", Cytometry Part B, 74B, (Suppl. 1) ,pp. 1-3, 2008.  World Health Organization, "Towards universal access: scaling up priority HIV/AIDS interventions in the health sector: progress report 2009" , 2009  World Health Organization, http://www.who.int/mediacentre/factsheets/fs]04/en/index.html,accessed July 25, 2010.  Yager, P., Domingo, G., Gerdes, J., 2008, "Point-of-Care Diagnostics for Global Health", Annu. Rev. Biomed. Eng. 2008, 10:107-44.  Mandy, F., Janossy, G. , Pilon, R., Faucher, S., "Affordable CD4 T-Cell Enumeration for Resource-Limited Regions: A Status Report for 2008. Cytometry Part B 2008; 74B (Suppl. 1): S27-S39, 2008.  Peter, T., Badrichani, A., Wu, E., Freeman, R., Ncube B, Ariki, F., Daily, J., Shimada, Y., Murtagh, M., 2008, "Challenges in implementing CD4 testing in resource-limited settings", Cytometry Part B 2008; 74B (Suppl. 1):S123-S130.  Katoh, T., Tokuno, R., Zhang, Y., 2008, "Micro injection molding for mass production using LIGA mold inserts", Microsystems Technology, 14:1507-1514.  Liu, J., Qiao, H., Liu, C., Xu, Z., 2009, "Plasma assisted thermal bonding for PMMA microfluidic chips with integrated metal microelectrodes", Sensors and Actuators B 141, pp. 646-65 1.  Ahn, C., Choi, J., Beaucage, G., 2004, "Disposable Smart Lab on a Chip for Pointof-Care Clinical Diagnostics", Proceeding of the IEEE, Vol., 92, No. 1,  Hale, M., 2009, "Development of a Low-Cost, Rapid-Cycle Hot Embossing System for Microscale Parts", Department of Mechanical Engineering, Massachusetts Institute of Technology, June 2009.  Attia, M., Marson, S., Alcock, J., 2009, "Micro-Injection molding of polymer microfluidic devices", Microfluid Nanofluid 7, pp 1-28.  Linares, R., 2010 " Manufacturability of Lab on a Chip Devices: Tolerance Analysis and Requirements Establishment", M.Eng. Thesis, Massachusetts Institute of Technology, Cambridge, MA.  Chung, X., Liu, Y., Irimia, D., Dixon, M., Demirci, U., Yang, L., Zamir, L., Tompkins, R., Rodriguez, W., Toner, M., Bashir, R., 2007, "Cell detection and counting trough cell lysate impedance spectroscopy in microfluidic devices", Lab Chip, 2007, 7, 746-755.  Chung, X., Irimia, D., Dixon, M., Sekine, K., Demirci, U., Zamir, L., Tompkins, R., Rodriguez, W., Toner, M., 2006 "A microfluidic device for practical label-free CD4+ T cell counting of HIV-infected subjects", Lab Chip, 2007, 7, 170-178.  Chung, X., Irimia, D., Dixon, M., Ziperstein, J., Demirci, U., Zamir, L., Tompkins, R., Toner, M., Rodriguez, W., 2007, "A Microchip Approach for Practical LabelFree CD4+ T-Cell Counting of HIV-Infected Subjects in Resource-Poor Settings" Journal of Acquired Immune Deficiency Syndrome, Volume 45, Number 3, July 1, 2007, pp 251-261.  Nguyen, N.-T., S. T. Wereley, "Fundamentals and Applications of Microfluidics", Boston, MA, Artech House, 2002.  Pranov, H., Koblitz, H., 2006, "On the Injection Molding of Nanostructured Polymer Surfaces", Polymer Engineering and Science, Feb 2006, 46, 2, Science Module pg. 46.  Deshpande, S., 1996, "Enzyme Immunoassays: From Concept to Product Development", Chapman & Hall, Chapter 7, pp 216 - 223.  Mooney, C. Z., 1997, Monte Carlo Simulation, Sage Publications, Thousand Oaks, California, US, Chap. 1.  Breedis, J. B., 2001, "Monte Carlo Analysis of a Passively Aligned Silicon Waferboard Package", 2001 Electronic Components and Technology Conference.  Lin, C., 1987, "A Statistical Analysis Tool for Variation Simulation Modeling", Computers Ind. Enggn, 13(1-4), pp. 386-391.  Hu, M., Lin, Z., Lai, X., Ni, J., 2001, "Simulation and analysis of assembly processes considering compliant, non-ideal parts and tooling variations", Intl Journal of Machine Tools & Manufacture, 41, 2233-2243.  Yu, M., Du, R., 2005, "Sensitivity Analysis of the Sheet Metal Stamping Processes based on Inverse Finite Element Modeling and Monte Carlo Simulation", AIP Conf. Proc. 778, pp. 813-818.  Wolfram Mathworld, http://mathworld.wolfram.com/Circle-CircleIntersection.html, accessed August 02, 2010.  Hornberg, A., 2006, Handbook of Machine Vision, Wiley-VCH, Weinheim, Germany.  Janossy, G., Mandy, F., O'Gorman, M., 2008, "Diagnostics in the Shadow of HIV Epidemics", Cytometry Part B, 74B, (Suppl. 1) ,pp. 1-3, 2008.