Hardware Reference
In-Depth Information
as the results derived by a traditional bench-top analyzer used by a laboratory
technician [
19
].
2.3
Reliability-Driven Error-Recovery
2.3.1
Error Recovery Strategies
In this subsection, we formulate the principles underlying error-recovery. For the
given bioassay protocol, we use the sensing system on-chip to evaluate the quality
of output droplets of each dispensing, mixing, dilution and splitting operation.
It is important to note that, the time cost for adding detection operations is
negligible, as the response time of on-chip sensors are in the scale of picoseconds
or nanoseconds [
13
].
One a microfluidic biochip, there are two categories of fluid-handling operations:
reversible and nonreversible operations. Reversible operations include dispensing
and splitting operations; nonreversible operations include mixing and dilution
operations. For errors that occur at reversible operations, their recovery processes
are relative simple. In a splitting operation, if two droplets with unbalanced volumes
are generated, then the biochip will first merge the two abnormal droplets to a larger
one and then split the larger droplet again. For errors that occur at a dispensing
operation, the chip can send the abnormal droplet back to the corresponding
reservoir and dispense another droplet. Thus for errors that occur at reversible
operations, the time cost for recovery is low and no additional droplets need to be
consumed.
The error-recovery process for nonreversible operations is more complicated.
In order to re-execute the corresponding nonreversible operations to correct the
error, we also need input droplets from operations whose outputs feed the inputs
of the failed operation. Thus we may need to re-execute all the predecessors of
the erroneous operation. For instance, if an error occurs at operation 7 in Fig.
2.6
a,
operations 1, 2, 3, 4, 5, and 6 may need to be re-executed. Thus the time cost for
executing error-recovery operations can be extremely high. The following strategies
are taken in our approach to reduce the incidence of the worst case:
•
For a splitting operation, if only one of its output droplets is used as the input
for the immediate successors, the other (redundant) droplet will be stored as a
backup for possible error-recovery at a subsequent stage. For example, operation
7inFig.
2.6
a is a splitting operation and it generates two output droplets. Only
one of these two droplets is used as the input of operation 9. (Note here each
circle in the sequencing graph stands for a fluid-handling operation. The unused
droplets are not shown in the sequencing graph.) If an error occurs at operation
9, the redundant droplet will be used as an input for re-execution.
