Lab-on-a-Silicon Chip (LOSC)

• Lab-on-a-Silicon Chip (LOSC) is a miniaturized analytical system that integrates microfluidics, sensing elements, and readout functions on silicon-based substrates.
• It employs diverse transduction methods—including optical, electrical, thermal, and impedance sensing—to enable applications such as immunoassays, antibiotic susceptibility testing, and digital microfluidics.
• Current research focuses on advancing monolithic integration, enhancing multiplexing, and reducing peripheral instrumentation to achieve cost-effective, fully integrated chip-scale diagnostics.

Lab-on-a-Silicon Chip (LOSC) denotes a class of miniaturized analytical systems in which microfluidics, sensing elements, and part of the readout chain are integrated on silicon or silicon-compatible substrates such as SOI, silicon-glass, Si/quartz, SiN, and Ge-on-Si. In the literature, LOSC includes label-free refractometric biosensors, all-electrical metabolic assays, integrated flow metrology, digital-microfluidic diagnostics, and chip-scale imaging. Representative realizations include silicon nanoresonator immunoassays for Prostate Specific Antigen in human serum (Yavas et al., 2017), a monolithic micro-ring-resonator plus on-chip spectrometer on SOI (Yoo et al., 2022), a membrane-free thermal flow sensor integrated into silicon-glass microfluidics (Ryzhkov et al., 2022), and a silicon-nanowire-FET platform for antibiotic susceptibility testing with a sample-to-results time within 20 minutes (Xu et al., 18 Aug 2025). The literature therefore treats LOSC less as a single device type than as a silicon-centric integration strategy.

1. Silicon material systems and fabrication routes

LOSC implementations span several silicon-based material stacks. The all-dielectric biosensing platform uses a quartz wafer coated with 50 nm of polycrystalline silicon, patterned into circular silicon nanodisks of height $h = 50 \,\mathrm{nm}$ and radii $r = 120, 140, 160,$ and $180 \,\mathrm{nm}$ by spin-coating negative e-beam resist, electron-beam lithography, development in salty developer, reactive-ion etching, resist strip, and $\mathrm{O_2}$ plasma cleaning (Yavas et al., 2017). The membrane-free thermal flow sensor is built on a 4″, 525 $\mu\mathrm{m}$ thick, double-side polished $<100>$ Si wafer with 4 $\mu\mathrm{m}$ wet $\mathrm{SiO_2}$ on both sides, bonded to a 4″, 150 $\mu\mathrm{m}$ thick Borofloat 33 glass wafer after DRIE of 50 $\mu\mathrm{m}$ deep channels and through-wafer vias, Ni lift-off on glass, and thin-film $r = 120, 140, 160,$0 passivation (Ryzhkov et al., 2022). Xu et al. implement the AST LOSC on SOI with a 55 nm top Si layer, a 145 nm buried oxide, 100 nm wide and 1.6 $r = 120, 140, 160,$1 long nanowires, 4 nm $r = 120, 140, 160,$2 by ALD, and an on-chip Ag/AgCl pseudo-reference electrode (Xu et al., 18 Aug 2025).

Photonic LOSC variants use standard integrated-photonics stacks. The MRR-SHFTS biosensor employs 220 nm Si on 3 $r = 120, 140, 160,$3 buried oxide on Si handle, with a single-step electron-beam lithography and chlorine-based ICP-RIE process defining the 500 nm wide waveguides, the ring resonator, the MZI arms, and the focusing sub-wavelength grating couplers (Yoo et al., 2022). The hybrid SiN/QD microdisk laser platform deposits a passive 250 nm SiN bus waveguide, planarizes with $r = 120, 140, 160,$4, and then forms a SiN/QD/SiN active sandwich with 100 nm bottom SiN, a $r = 120, 140, 160,$5 nm CdSe/CdS colloidal QD film, and 100 nm top SiN; all steps are performed at $r = 120, 140, 160,$6 and the reported yield exceeds 90% across a 4-inch silicon wafer (Xie et al., 2019). For mid-infrared LOSC, Torres-Cubillo et al. compare several thin-film waveguides and identify Ge-on-Si as especially suitable because it is fully CMOS-compatible and relies on mature Si/Ge processing (Torres-Cubillo et al., 2024).

A recurrent feature across these reports is co-fabrication of sensing and fluidic functionality within one technological flow, or at least within compatible flows that preserve alignment and packaging simplicity.

Platform	Representative stack	Key process sequence
Silicon nanoresonator biosensor	quartz + 50 nm polycrystalline Si	HSQ e-beam lithography, RIE, $r = 120, 140, 160,$7 plasma, PDMS integration
Silicon-glass MTFS	Si with 4 $r = 120, 140, 160,$8 wet $r = 120, 140, 160,$9 + Borofloat 33	DRIE, Ni lift-off, $180 \,\mathrm{nm}$0 passivation, anodic bonding
SiNWFET AST	SOI + 4 nm $180 \,\mathrm{nm}$1 + on-chip Ag/AgCl	EBL/RIE, As implant, NiSi silicidation, ALD, plasma bonding
SOI photonic biosensor	220 nm Si on 3 $180 \,\mathrm{nm}$2 BOX	single-step e-beam lithography, ICP-RIE

2. Microfluidic orchestration and sample handling

Microfluidics in LOSC is not merely a packaging layer; it defines sample delivery, timing, concentration, and dead-volume budgets. The silicon nanoresonator biosensor integrates PDMS multilayer soft-lithography with 8 parallel microfluidic channels and Quake-style microvalves. The flow layer contains 100 $180 \,\mathrm{nm}$3 deep channels bonded to the Si/quartz sample, while electronic pressure controllers operating from 0 to 40 psi actuate the valves under software control. The reported scheme supports parallel assay execution across capture-antibody, sample, control, rinse, and detection-antibody steps, while allowing real-time optical interrogation by transmission microscopy coupled to a VIS-NIR spectrometer (Yavas et al., 2017).

The membrane-free thermal flow sensor is specifically motivated by a common LOSC bottleneck: external flow meters add dead volume through tubing and are not usually fabricated in the same technological cycle as the microchannels. Its sensing elements sit on the glass surface and are never wetted by the fluid, while the channel is etched in the opposing Si wafer. Because the sensor is in-plane with the channel and the inlet/outlet use through-silicon vias or 1.15 mm EPDM O-rings directly at the chip edges, the design has virtually no extra tubing or external manifold volume. The report states that multiple MTFS units can be laid out on a single silicon-glass chip for true multichannel integration, and that dead volume can be reduced by factors of 10–100 compared to off-chip meters (Ryzhkov et al., 2022).

The AST LOSC uses microfluidics not only for transport but also for bacterial concentration. PDMS channels of 100 $180 \,\mathrm{nm}$4 width and 20 $180 \,\mathrm{nm}$5 height feed 35 pL culture chambers, each containing one SiNWFET and a paired AgCl/Ag pseudo-reference electrode. The inlet channels are 2 $180 \,\mathrm{nm}$6 wide and 1.5 $180 \,\mathrm{nm}$7 high, allowing E. coli passage, whereas 0.5 $180 \,\mathrm{nm}$8 exit channels block bacterial escape. Two loading modes are defined: bypass loading for $180 \,\mathrm{nm}$9 cells/mL and forced loading for $\mathrm{O_2}$0 cells/mL, the latter improving capture efficiency by forcing all flow through the chambers (Xu et al., 18 Aug 2025).

Other LOSC-adjacent systems emphasize different forms of fluidic programmability. The optofluidic ptychography device uses a straight PDMS channel 5 mm wide and 0.5 mm tall, with either electrokinetic flow at approximately 36 V DC and 40–80 $\mathrm{O_2}$1 or pressure-driven flow at 100–400 $\mathrm{O_2}$2 (Song et al., 2021). Zhang et al.’s digital-microfluidics platform addresses droplets on a standard double-plate DMF chip with 169 hexagonal electrodes and up to 180 routed channels through HV507 multiplexers, reusing the actuation infrastructure for sensing (Zhang et al., 2019). A plausible implication is that LOSC fluidics now span continuous flow, droplet transport, valve-based routing, and flow-embedded concentration, rather than a single microchannel paradigm.

3. Transduction principles and analytical readout

LOSC transduction is heterogeneous. Optical, electrical, thermal, and impedance readouts coexist, and several reports explicitly seek to eliminate bulky off-chip instrumentation.

All-dielectric biosensing with silicon nanodisks relies on extinction resonances dominated by the collective electric-dipole Mie resonance of the array. The bulk refractive-index sensitivity is defined as

$\mathrm{O_2}$3

with $\mathrm{O_2}$4 obtained from the centroid shift of the extinction peak, and the figure of merit is

$\mathrm{O_2}$5

For the optimal array with $\mathrm{O_2}$6 and $\mathrm{O_2}$7, the reported values are $\mathrm{O_2}$8, $\mathrm{O_2}$9, and $\mu\mathrm{m}$0. FEM simulations show that the near-field $\mu\mathrm{m}$1 is concentrated at the disk edges and decays over $\mu\mathrm{m}$2 from the surface, which the paper identifies as advantageous for multilayer immunoassays (Yavas et al., 2017).

The SOI MRR-SHFTS platform uses a symmetric add-drop micro-ring resonator as the sensing element and a spatial-heterodyne Fourier-transform spectrometer built from $\mu\mathrm{m}$3 unbalanced MZIs as the readout engine. The ring obeys the resonance condition $\mu\mathrm{m}$4, with reported $\mu\mathrm{m}$5 and $\mu\mathrm{m}$6. The bulk sensitivity is $\mu\mathrm{m}$7, while the integrated spectrometer yields spectral resolution $\mu\mathrm{m}$8 over a bandwidth of $\mu\mathrm{m}$9, leading to a reported limit of detection of $<100>$0 (Yoo et al., 2022).

Electrical LOSC transduction is exemplified by the SiNWFET AST platform. Live bacteria metabolize glucose and secrete organic acids, increasing $<100>$1 concentration in unbuffered medium. The resulting surface-potential shift at the $<100>$2 gate dielectric produces a threshold shift

$<100>$3

while in subthreshold the device follows

$<100>$4

The reported pH sensitivity is near-Nernstian, $<100>$5 at $<100>$6, and the $<100>$7 signal is linearly resolvable with SD $<100>$8 (Xu et al., 18 Aug 2025).

The membrane-free thermal flow sensor operates in constant-heater-temperature calorimetric mode. A heater and four matched thermoresistors are patterned on glass above the channel, and the upstream/downstream temperature asymmetry provides the flow signal. With a Wheatstone bridge and small bias current $<100>$9, the output is

$\mu\mathrm{m}$0

Finite-element optimization in COMSOL gives an optimal sensor-to-heater distance $\mu\mathrm{m}$1 in the 250–500 $\mu\mathrm{m}$2 range, and the fabricated device measures 2–30 $\mu\mathrm{m}$3 with relative flow error below 5% and sub-second response (Ryzhkov et al., 2022).

Zhang et al.’s DMF platform uses impedance sensing for both connectivity diagnostics and droplet localization. The basic electrical model is

$\mu\mathrm{m}$4

A 0–280 V square wave at approximately 12 kHz is applied to a selected electrode, the MCU samples at 200 kHz for 256 points, and an in-MCU FFT extracts the magnitude $\mu\mathrm{m}$5 in less than 2 ms per channel. Typical thresholds are reported as $\mu\mathrm{m}$6 for open circuit, $\mu\mathrm{m}$7 for oil/air, and $\mu\mathrm{m}$8 for a DI-water droplet (Zhang et al., 2019).

Two additional modalities broaden the LOSC envelope. Optofluidic ptychography places a microfluidic channel on a coverslip whose opposite side is coated with a sparse monolayer of 200 nm polystyrene beads acting as a scattering layer; the assembly is placed on a CMOS image sensor, and a modified ePIE reconstruction recovers quantitative complex images from diffraction data (Song et al., 2021). Microfluidic quantum sensing with NV centers in diamond, although demonstrated on fused silica, is stated to be transferable in principle to silicon-based substrates with suitable surface passivation and bonding chemistries. It combines microfluidics with the NV spin Hamiltonian

$\mu\mathrm{m}$9

supporting relaxometry, nanoscale NV-NMR, and microscale NV-NMR (Allert et al., 2022).

For mid-infrared LOSC, the design problem is formalized through the evanescent-field confinement factor $\mathrm{SiO_2}$0, the propagation loss $\mathrm{SiO_2}$1, and the limit of detection

$\mathrm{SiO_2}$2

with optimum interaction length $\mathrm{SiO_2}$3 and waveguide figure of merit $\mathrm{SiO_2}$4. In the comparison at $\mathrm{SiO_2}$5, Ge-on-Si achieves the highest reported FOM, 4.46 $\mathrm{SiO_2}$6 (Torres-Cubillo et al., 2024).

Modality	Primary observable	Representative figures
Silicon nanoresonator biosensing	extinction peak shift $\mathrm{SiO_2}$7	$\mathrm{SiO_2}$8, $\mathrm{SiO_2}$9
MRR + SHFTS	resonance shift reconstructed on-chip	$\mu\mathrm{m}$0, $\mu\mathrm{m}$1
SiNWFET AST	$\mu\mathrm{m}$2 from metabolic pH	$\mu\mathrm{m}$3, SD $\mu\mathrm{m}$4
MTFS	bridge voltage from $\mu\mathrm{m}$5	2–30 $\mu\mathrm{m}$6, relative error $\mu\mathrm{m}$7
DMF impedance sensing	FFT magnitude $\mu\mathrm{m}$8 at $\mu\mathrm{m}$9	$\mu\mathrm{m}$0, 300 ms for 169 channels
Optofluidic ptychography	reconstructed complex image	550 nm linewidth, 30 fps
NV microfluidic sensing	$\mu\mathrm{m}$1, NV-NMR, magnetometry	$\mu\mathrm{m}$2 Gd$\mu\mathrm{m}$3, $\mu\mathrm{m}$4 Hz linewidth

4. Representative assay and imaging workflows

The LOSC literature contains complete analytical workflows rather than isolated sensor calibrations. In the silicon nanoresonator PSA assay, anti-PSA capture antibody is immobilized by passive adsorption through a 20 $\mu\mathrm{m}$5 PBS solution for 30 min, followed by a PBS rinse. PSA in PBS + 1% BSA is then flowed for approximately 10–15 min, followed by another rinse, then an anti-PSA detection antibody for approximately 10 min, and finally a PBS rinse and spectral readout. The total assay time is approximately 1 h. The reported calibration uses a 4-parameter logistic function,

$\mu\mathrm{m}$6

and for the optimal sensors gives $\mu\mathrm{m}$7 values of 0.69 ng/mL and 0.74 ng/mL in buffer, with intra-chip RSD of 0.5%–5.1%. In 50% human serum, the reported $\mu\mathrm{m}$8 is 1.6 ng/mL, below the clinical cut-off of 4 ng/mL (Yavas et al., 2017).

Xu et al.’s AST LOSC replaces growth-based culture timing with metabolic monitoring in confined chambers. Starting samples as low as $\mu\mathrm{m}$9 can be loaded in forced-loading mode, yielding approximately 2–4 cells per chamber in 8 min and a concentrated chamber density of approximately $r = 120, 140, 160,$00. The workflow then uses 1 min for media exchange, 1 min for FC-70 oil sealing, and electrical readout of metabolic acidification. In unbuffered 0.9% NaCl + 1% glucose, the report gives a 60 mV $r = 120, 140, 160,$01 in 20 min, versus 50 min in LB. The AST decision criterion is a $r = 120, 140, 160,$02 difference of at least 15 mV between treated and control chambers. For a clinical UPEC isolate, the total sample-to-AST result is reported as $r = 120, 140, 160,$03 min for $r = 120, 140, 160,$04–$r = 120, 140, 160,$05 samples, with 24 parallel culture chambers per chip and two independent fluidic arrays, giving 48 tests per chip (Xu et al., 18 Aug 2025).

Digital-microfluidic LOSC diagnostics can include a hardware-verification step before fluidic interpretation. In Zhang et al., each electrode is addressed through high-voltage multiplexers, sampled for approximately 1.28 ms, and classified by FFT magnitude as open, medium-filled, or droplet-occupied. Connectivity tests showed zero false positives when half or all FPCs were cut, and full-array scans of 169 channels took approximately 300 ms, enabling real-time droplet tracking at a few Hz update rate (Zhang et al., 2019).

The optofluidic ptychography workflow is imaging-centered rather than endpoint-centered. Raw diffraction frames are averaged to initialize the scattering layer estimate, object motion is tracked by back-propagation and registration in the channel plane, and a modified ePIE jointly updates the object wavefront and the scattering layer over 3–5 iterations on 90–500 frames. Using a Sony IMX226 monochrome CMOS sensor with 1.85 $r = 120, 140, 160,$06 pixels and 532 nm illumination, the device resolves a 550 nm linewidth and achieves a ptychographic video frame rate of 30 fps. Reported specimens include C. elegans, yeast cells, paramecium, and closterium sp. (Song et al., 2021).

The quantum microfluidic platform extends LOSC-style workflows to spin-based chemical analysis. Paramagnetic-ion detection uses $r = 120, 140, 160,$07 relaxometry with delay sweeps from 200 ns to 5.5 ms in 51 log-spaced points, reporting a decrease in $r = 120, 140, 160,$08 from approximately 2 ms in pure water to approximately 0.5 ms at 10 $r = 120, 140, 160,$09 Gd$r = 120, 140, 160,$10 and a detection limit of approximately 1 $r = 120, 140, 160,$11 in a 600 nL channel. Microscale NV-NMR uses a 50 $r = 120, 140, 160,$12 thick NV layer, a 45 $r = 120, 140, 160,$13 laser spot, and a detection volume of approximately 130 pL, with reported spectral resolution $r = 120, 140, 160,$14 Hz and resolved $r = 120, 140, 160,$15 Hz in trimethyl phosphate (Allert et al., 2022). Because this implementation is on fused silica rather than silicon, its relevance to LOSC is prospective rather than canonical; the paper explicitly states that transfer to silicon-based substrates is possible in principle.

5. Multiplexing, monolithic integration, and active photonics

A defining LOSC ambition is to move from single-sensor chips to parallel, partially self-contained analytical microsystems. Multiplexing is already explicit in several reports: the silicon nanoresonator arrays occupy approximately $r = 120, 140, 160,$16 footprints and allow up to 32 individually addressed regions in the optical field of view, while the microfluidic overlay provides 8 parallel channels (Yavas et al., 2017). The AST LOSC provides 24 chambers per chip and two fluidic arrays for 48 parallel tests (Xu et al., 18 Aug 2025). The DMF impedance platform demonstrates 180 addressable channels and full-array scans of 169 electrodes (Zhang et al., 2019). The integrated SHFTS uses 32 unbalanced MZIs as a detector-side multiplexing structure (Yoo et al., 2022). The MTFS work states that multiple sensing units can be laid out on a single silicon-glass chip, and that the footprint per sensor is below $r = 120, 140, 160,$17 and can shrink to $r = 120, 140, 160,$18 (Ryzhkov et al., 2022).

Monolithic readout is a recurrent goal. The MRR-SHFTS platform is explicitly designed to eliminate the external optical spectrum analyzer by reconstructing the MRR spectrum on-chip (Yoo et al., 2022). The all-dielectric nanoresonator paper notes that a single-wavelength LED or a small-footprint broadband source plus a mini-spectrometer suffice for readout, and that on-chip photodetectors are a future path to fully monolithic sensing (Yavas et al., 2017). Xu et al. emphasize all-electrical readout and estimate low per-chip cost below several USD in volume, while noting that the present readout instrument is still bulky and should be replaced by CMOS electronics (Xu et al., 18 Aug 2025).

Active on-chip photonic sources are represented by the hybrid QD/SiN microdisk laser. The reported device couples a 7 $r = 120, 140, 160,$19 diameter SiN/QD/SiN microdisk to a passive SiN bus waveguide, achieves a threshold pump fluence of $r = 120, 140, 160,$20, and shows no measurable degradation over weeks with wavelength drift below 1 nm (Xie et al., 2019). The paper explicitly frames this approach as relevant to optical communication, lab-on-a-chip, and gas sensing, and its integration with passive SiN waveguides makes it pertinent to LOSC architectures that currently depend on external illumination.

Mid-infrared LOSC points toward a more system-level optimization framework. In the comparative thin-film analysis at $r = 120, 140, 160,$21, the Ge-on-Si platform is reported with $r = 120, 140, 160,$22, $r = 120, 140, 160,$23, $r = 120, 140, 160,$24, and a paper-reported optimum interaction length of 0.54 mm via the full LoD curve, with $r = 120, 140, 160,$25. The same study recommends flip-chip QCL arrays, GaSb QCL-Si photonics, Ge or InGaAsSb photodiodes, and SiO$r = 120, 140, 160,$26/SiN or silicon-glass microfluidics as integration routes (Torres-Cubillo et al., 2024). This suggests that LOSC is converging toward complete spectroscopic subsystems rather than isolated sensors.

6. Constraints, misconceptions, and research directions

Several misconceptions are corrected by the literature. First, LOSC is not synonymous with optical biosensing. The corpus includes refractometric nanophotonics (Yavas et al., 2017), integrated optical spectroscopy (Yoo et al., 2022), thermal flow metrology (Ryzhkov et al., 2022), impedance-based connectivity and droplet sensing (Zhang et al., 2019), electrical phenotypic microbiology (Xu et al., 18 Aug 2025), computational imaging (Song et al., 2021), and spin-based chemical analysis (Allert et al., 2022). Second, chip-scale integration does not imply that all peripherals have disappeared. The nanoresonator PSA platform still uses transmission microscopy and a VIS-NIR spectrometer (Yavas et al., 2017); optofluidic ptychography uses a laser and GPU/FPGA-based reconstruction (Song et al., 2021); the quantum platform requires 532 nm optical excitation and microwave control near 1.95 GHz (Allert et al., 2022); and the AST platform notes that its present HP 4155A-based readout remains bulky (Xu et al., 18 Aug 2025). What has changed is the location of the critical sensing interface: it is brought onto the silicon chip.

Current limitations are modality-specific. The dielectric nanoresonator platform identifies sharper resonances, site-directed antibody immobilization, reduction of non-specific binding, and bound states in the continuum as routes to improved performance (Yavas et al., 2017). The AST LOSC relies on acid-producing metabolism, so non-acidic pathogens may require alternative readouts, and chamber loading time increases at very low bacterial counts (Xu et al., 18 Aug 2025). The MRR-SHFTS LoD is constrained by a spectrometer resolution of approximately 3.1 nm (Yoo et al., 2022). In mid-IR waveguides, the trade-off between evanescent overlap and water absorption explicitly limits usable path length, hence the emphasis on $r = 120, 140, 160,$27 optimization (Torres-Cubillo et al., 2024). The membrane-free MTFS, while robust and chemically tolerant, is still one element in a larger control loop that must include pressure sources and calibration (Ryzhkov et al., 2022).

Research directions in the cited work consistently point toward tighter hardware closure. The DMF platform proposes an on-chip ASIC front-end, phase extraction, and swept-frequency dielectric spectroscopy (Zhang et al., 2019). The AST platform proposes single-cell AST, CMOS multiplexers, low-power mixed-signal front ends, and on-chip valves for fully automated sample-to-answer operation (Xu et al., 18 Aug 2025). The quantum microfluidic platform points toward monolithic silicon CMOS integration through bonded diamond platelets, co-fabricated CPWs, and readout photodiodes (Allert et al., 2022). The optofluidic ptychography work suggests multimodal extension through fluorescence or spectral channels (Song et al., 2021). The microdisk-laser platform suggests combining on-chip laser, detector, interferometer, and microfluidics in a single fabrication platform (Xie et al., 2019).

Taken together, the literature presents LOSC as a progressively expanding integration domain in which silicon serves not only as a substrate but as a fabrication, packaging, and systems-integration framework. The central technical problem is no longer whether sensing can be miniaturized onto chip-scale structures, but how far sample handling, excitation, signal conditioning, spectrometry, and decision logic can be collapsed into a single silicon-compatible microsystem.