Source: Biosensors, by Hyunwook Ryu; Won-Chul Lee; Jongwon Lee (August 16, 2026). AI-generated summary by biochip.com, published . Not independently reviewed.
Key takeaways
- Ryu, Lee, and Lee review memristive biosensors that combine sensing, memory, and computation to reduce data movement.
- The field is shifting from direct liquid-exposed memristors toward indirect architectures that electrically connect protected devices to sensing electrodes.
- CMOS compatibility, device variability, and reliable multi-threshold sensing remain barriers to practical electrochemical biosensing systems.
A review by Hyunwook Ryu, Won-Chul Lee, and Jongwon Lee examines how memristive devices could reshape electrochemical biosensors by moving data processing closer to where biological signals are collected. A memristor is an electronic component whose electrical resistance changes with its history of applied voltage or current, allowing it to retain a memory-like state without continuous power. That property makes these devices candidates for biosensors that can sense, store, and interpret a signal on a compact platform. The review traces a shift from early designs that exposed a memristor directly to a biological liquid toward architectures that protect the device while preserving its ability to process sensor outputs. It also follows a broader systems transition from off-chip analysis, where raw signals travel to a separate processor, to on-chip designs that combine sensing and computation. The destination is in-sensor computing, an approach in which sensing, memory, and decision-making share the same physical location. Such arrangements could cut the energy and delay associated with moving continuous streams of raw biological data through conventional electronics. But the review also makes clear that compatibility with standard semiconductor manufacturing, variation between devices, and dependable multi-threshold operation remain important barriers to practical systems.
Why data movement is the problem
Most electrochemical biosensors still follow a familiar division of labor: a sensor detects a chemical event, then sends an analog electrical signal to separate circuitry for conversion and analysis. It is a bit like shipping every grocery item to a distant kitchen before deciding what meal to make. The physical trip between sensor, memory, and processor adds delay and consumes power.
This separation becomes more troublesome when a device must monitor biology continuously or operate at the point of care, meaning close to a patient rather than in a centralized laboratory. Raw sensor data can arrive as an unbroken stream, even though only a small fraction may indicate a meaningful event. Sending all of it to an external processor can burden battery-powered and compact diagnostic hardware.
What a memristive biosensor adds
Memristive devices offer a different hardware model because their electrical conductance, or ease of current flow, can be changed and retained. In practical terms, they can act partly like memory and partly like a switch or computational element. Their non-volatile behavior means the programmed state can remain after power is removed, while low-voltage switching and small physical dimensions make them appealing for dense electronic systems.
For biosensing, this behavior can turn a changing chemical signal into an electrical state that holds useful information. Rather than merely collecting a measurement for another chip to interpret, the memristive element can participate in classifying whether a signal has crossed a predefined condition. The review frames this as a route toward compact diagnostic platforms that reduce the need to shuttle data between separate components.
From direct contact to protected architectures
Early memristive biosensor configurations often used direct sensing. In these designs, the active memristive layer contacted the biological solution itself, allowing ions in the liquid to influence the device's switching behavior. This can provide a close coupling between chemistry and electronics, but it leaves the active material exposed to a demanding environment.
Complex electrolytes can chemically degrade materials or destabilize their structure during repeated use. That creates a basic engineering tradeoff: direct exposure may support sensitive transduction, the conversion of a biological event into an electrical output, but it can compromise long-term reliability. A biosensor intended for routine or repeated measurements needs both responsive sensing and stable operation.
The review identifies a move toward indirect sensing as a response. Here, a primary sensing electrode interacts with the biological sample, while the memristive device remains physically separated from the liquid and electrically connected to the sensor output. Think of the arrangement as putting a durable control room behind a window rather than placing the control equipment directly in a corrosive factory floor.
Computing at the sensing interface
Indirect architectures do more than shield a sensitive component. They allow the sensor electrode to generate an input that the memristive device can convert into a stored state or a decision signal. The core device therefore remains part of the sensing system without being the surface that must survive direct exposure to biological fluids.
The next step is fully integrated on-chip implementation, where sensing and electronic functions are assembled on the same chip rather than distributed across separate boards or external processors. This matters because long electrical connections and repeated analog-to-digital transfers can consume energy. Keeping the work local supports what engineers call data-stationary processing: data stay near the place where they were generated.
In-sensor computing pushes that idea further by co-locating sensing, memory, and computation within a single physical platform. Instead of recording every raw voltage trace and analyzing it later, a device could apply threshold-based logic at the sensor. A threshold is simply a chosen cutoff, similar to a smoke alarm that stays quiet until particles in the air rise above a defined level.
Threshold decisions and replaceable electrodes
The review highlights a trajectory toward fully decoupled nodes that use deterministic threshold sensing-based decision logic. Deterministic means the system produces a defined result when an input satisfies a specified condition, rather than relying on an uncertain interpretation of every raw signal. This can be useful for tasks that require an immediate yes-or-no or category-level response.
These proposed arrangements can use replaceable sensing electrodes while keeping the memristive processing element protected. That design separates the disposable part that contacts a sample from the more durable electronic part that stores and processes information. It aims to avoid material degradation in the core device without giving up local computation at the transducer level.
Why This Matters
For point-of-care testing, power and physical size are not secondary details. A useful device may need to operate outside a major laboratory, handle samples near the patient, and deliver an interpretable result without relying on continuous communication with a distant computer. Reducing data movement could help make those constraints easier to meet.
The concept also speaks to edge artificial intelligence, or edge AI, in which computation occurs near the source of data rather than in a remote data center. In a biosensor, the relevant edge is the sensing interface itself. Memristive hardware could provide a physical means to make simple decisions where the biological signal first appears.
What must be solved next
Several obstacles separate this architectural promise from dependable deployment. The devices must be compatible with complementary metal-oxide-semiconductor, or CMOS, technology, the manufacturing platform used for most modern electronic chips. They must also control device-to-device variability, because nominally identical memristors can switch differently from one another.
Reliable multi-threshold sensing is another challenge. A system that distinguishes among several signal levels must set and retain those boundaries consistently, even as devices age or experience changing conditions. Progress will depend on designs that combine protected sensing interfaces, stable memristive behavior, and circuit-level integration without sacrificing the energy advantages that motivate in-sensor computing in the first place.