CLCAD: Cross-Layer Collaborative Anomaly Detection for Continuous-Flow Microfluidic Biochips

Abstract

Continuous-flow microfluidic biochips are widely used in biochemical applications because of their precise control, high integration, and automation. During bioassay execution, the control layer applies pneumatic signals to valves according to the valve activation sequence, ensuring reliable execution in the flow layer. However, in an insecure supply chain, attackers can induce anomalous valve states through Trojan insertion or other malicious means, causing anomalous execution. For example, attackers can tamper with the valve activation sequence of the flow layer or the design scheme of the control layer, thereby causing incorrect valve activation or disrupting control signal delivery to the target valves, respectively. To detect anomalies during execution, we propose CLCAD, a cross-layer collaborative anomaly detection method that collaboratively identifies anomalies in valve states caused by potential attacks in both the flow and control layers. To address tampering with the valve activation sequence, we propose a weight calculation strategy based on fluid scheduling and a checkpoint selection strategy based on discrete particle swarm optimization, enabling checkpoint configuration optimization and detection capability improvement under limited detection resources. To address tampering with the control layer design scheme, we propose a valve addressing scheme detection strategy, a valve synchronization requirement detection strategy, and a test pattern selection strategy based on discrete social learning particle swarm optimization, enabling the selected test patterns to cover a wider range of anomalous behaviors. Experimental results demonstrate that CLCAD effectively improves anomaly detection probability and return on investment while maintaining low detection cost across multiple benchmarks.