0740-7475/11/$31.00 © 2011 IEEE
Published by the IEEE Computer Society
All About Liquid Scan Chains— and More
It is Design & Test's policy to commission independent reviews of books written by the editor in chief. Krishnendu Chakrabarty was not involved in any phase of the commissioning or editing of this review.
—Robin Baldwin, Editorial Business Operations manager
This is a review of Digital Microfluidic Chips: Design Automation and Optimization, by Krishnendu Chakrabarty and Tao Xu.
We routinely talk about data and bits flowing through our chips. Apparently, there are other types of chips in which this flow is literal in that liquid droplets truly flow from one part of the chip to the other. Not surprisingly, these chips have design and test problems just like regular silicon chips.
A microfluidic biochip integrates biochemistry and electronics (these types of chips will find applicability in such industries as biotech and pharmaceuticals). A microfluidic biochip enables the analysis and mixing of tiny amounts of biological samples rapidly and automatically. The book under review, one of whose authors is Design & Test's editor in chief, describes the application of design automation and test techniques to these chips.
We begin with the question of why a person interested in the design and test of computers should be interested in this topic, except perhaps when you have labwork done. There are two reasons. First, this is going to be an increasingly important market, and knowing something about it might eventually be useful for your career. Second, it is intriguing to see how many of the concepts of optimization and test we've applied to digital devices are also useful for chips so different from what we know.
The first chapter, the introduction, defines what a microfluidic biochip is and outlines the problems the book will address. One innovation—to me at least—is Figure 1.6, which gives a graphical outline of the rest of the book. This is an interesting alternative to the boring, old outline section. I'd have liked a bit more of a tutorial in this chapter, and some indication here of the size of these chips. We do learn that later, but at this point I wondered how much of a problem biochip design really is.
Briefly, microfluidic biochips can be considered as a grid of cells, with a path through the chip such that no two droplets being routed will interfere with each other. Control of the route is done through chip pins, and there are constraints on how many of these are allowed. This presents us with several interesting problems.
The first problem is synthesis. Not only must we design a biochip to allow the droplets to be routed from their origin to where they are combined with others and then to their final destination, but the design should be structured better so that there is still a path for droplet routing even when some of the cells of the chip have defects.
The second chapter provides a means of doing routing-aware synthesis, which tries to reduce the need for expensive resynthesis to create routes after synthesis by using a routability estimation metric during synthesis. The chapter then moves on to a discussion of defect tolerance. Two ways to attain this are given. First, postsynthesis defect tolerance tries to reconfigure a defective chip to be functional. Methods are described for doing this locally, or only for cells that are intended to be visited by the droplet that's moving from cell to cell in the chip after the defect blocks the droplet, in order to save time versus reconfiguring the entire chip. In the second method, by using presynthesis defect tolerance, a biochip engineer tries to synthesize a design in order to increase the probability that such a reconfiguration will be found.
Moving droplets involves applying signals within the chip to open and close electrodes. As chips grow bigger, doing this by directly addressing them from the pins becomes expensive. Chapter 3, on pin-constrained chip design, describes ways to design chips when you have a limited number of pins while still allowing as many droplets to move in parallel as possible. This chapter applies some well-known algorithms and heuristics to the problem, which involves both scheduling and partitioning. The algorithm that's the most fun is one that maps control signals to electrodes in a partition based on a strategy for the Connect-5 game. This chapter is an excellent example of the observation many of us have made that nearly any problem can be solved by mapping it to an already-solved problem. Perhaps we need to search for solutions in more places than we normally consider (and perhaps I can take my games closet as a business expense!).
I've mentioned that microfluidic biochips have defects just like silicon chips. The effects of these are more like what you might see in your kitchen faucet than your typical ASIC. One defect causes electrodes blocking transport to be stuck on, and the result is a constant drip, where droplets reaching the stuck-on electrodes won't stop. How do we test for these?
The answer, covered in chapter 4, looks a lot like scan test, with droplets instead of bits moving through a chain. Just as with scan test, we can use a single chain or parallel chains. Unlike with scan test, though, you need to test both rows and columns, seeing the biochip as a 2D grid.
Since we are using fault tolerance, we need to identify fault sites in order to route around them. We also need to diagnose the faults we find, which can get tricky. Given that we have poor observability for droplets that disappear in the middle of the array, we can have multiple defects, and we can even have cells that are untestable because they cannot be reached because of other defects. Finally, scan-like testing is not enough. We must test certain capabilities of cells—such as splitting or combining droplets—functionally.
Wherever we find test, we also find design for testability. If the chip is designed using direct addressing, we can control every cell, so this is not a problem. For pin-constrained designs, though, we might have cases in which there is less than 100% testability, either structural or functional. "Structural testability" means the percentage of electrodes that can be traversed by a droplet during a test. "Functional testability" means the percentage of testable functional units, such as splitters. In this context, "DFT" means considering test as a function, and synthesizing the test functionality with the regular functionality, making the resulting chip easier to test.
The book concludes with the application of the techniques described in earlier chapters to a biochip for protein crystallization. The application description is useful in tying together the unfamiliar concepts presented in the preceding chapters and is an excellent way to summarize the book's contents.
Some general comments. The book is well written, and has many good examples and excellent diagrams. I would not recommend it as a survey. Although many references are given, the focus is on these newly proposed techniques, not the state of the field as a whole. There is no historical summary.
Should you read it? At just under 200 pages, including index, the book is a good short summary of the problems of the field, and one view of how they may be solved. If this area appeals to you, it might be a good starting point. It is still a new area, so those with expertise in similar problems might find this to be a good area of research. As for me, although I found the book to be very interesting, I don't plan on dropping my current job to get into the droplet pushing business anytime soon.