Hardware Reference
In-Depth Information
It is important to note that, for some operations, the recovery subroutines may
change depending on the error. For example, operation 7 in Fig.
2.6
a generates two
droplets; one of them is used in the subsequent operations and the other is stored on
chip as the “backup droplet”. If a single error occurs at operation 14, the biochip will
re-execute operations 8
10,and12. However, if an error occurs at a predecessor
of operation 14, the recovery subroutine for operation 14 will be different. For
example, when an error occurs at operation 9, the backup droplet of operation 7
will be used as the input for error-recovery. If another error occurs afterwards at
operation 14, there is no additional droplets available at the output of operation 7.
Thus the recovery subroutine of operation 14 has to be expanded and it will now
include operations 1
10,and12. Therefore the recovery steps are completely
different from the case when an error occurs at operation 14.
After the recovery subroutine of an operation is determined, the control software
will update the sequencing graph and the corresponding DAG, and then the dynamic
re-synthesis step will be performed.
2.4.2.3
Step 3: Dynamic Re-synthesis
In the cyberphysical system envisioned here, when an error is detected at a
checkpoint, it will trigger the generation of a new mapping of the remaining steps
(including proper handling of intermediate results) of the bioassay. This process is
referred to here as
re-synthesis
, on the basis of the initial design obtained from the a
priori synthesis step. The process of re-synthesis has the following requirements:
•
The interruption of other operations should be avoided. Consider the following
example. For the bioassay with synthesis results shown in Table
2.3
, suppose
an error-recovery process is triggered by an error in operation Mix 3 at time
instance 10. When the error-recovery process is triggered, operation Mix 1 is
being implemented. In order to avoid the interruption of Mix 1, in the re-synthesis
results, the schedule and resource assignment results for Mix 1 should be the
same as in the inial synthesis results.
•
The electrodes at which an error has been deemed to have occurred should be
bypassed in the re-synthesis results.
•
The completion time of the bioassay should be minimized.
To satisfy these requirements, we propose two re-synthesis strategies for dynamic
re-synthesis. The first strategy is based on a local greedy algorithm, and the second
strategy is a PRSA-based global optimization algorithm [
1
].
For the greedy algorithm, the first step is to determine all operations that must be
adjusted in the re-synthesis result. These operations include: the operations in the
error-recovery graph, the erroneous operation, and the set of subsequent operations
that will be implemented on electrodes with defects in the initial synthesis result.
Other operations will be executed based on the initial synthesis result.
Dynamic re-synthesis on the microfluidic array can be modeled as the
module
placement with obstacles
problem since the synthesis results for part of the
