The substrate question has always been the awkward part of surface-enhanced Raman. The optics are interesting, the chemistry is well understood, the enhancement numbers in the literature are spectacular - and yet the field has taken decades to deliver substrates that an industrial laboratory can buy in a thousand-piece lot, mount in a flow cell, and trust for a year of routine work. Over 2025 and the first half of 2026, the substrate literature has matured in a way that is worth tracking, because the gap between a single-molecule demonstration and a deployable consumable has narrowed by something more than zero, even if it has not closed.

This piece walks through what the most recent reviews and primary papers actually report. The framing is industrial: which directions look like they might produce a part number, and which remain fundamental science that happens to use Raman as a probe. For background on when SERS is the right tool at all, see our earlier explainer on where surface enhancement actually helps in process analytics. For the broader Raman picture, the inline Raman buyer’s guide sets the comparison point.

Where the 2025-2026 substrate work is concentrated

Two recent reviews provide a useful map. Tian and colleagues, writing in Battery Energy in 2025, organise the field by structural feature: micro- and nanostructure families, the materials used to build them, and the preparation routes that scale. A complementary review in Coordination Chemistry Reviews in 2025 takes flexible substrates as its subject and covers design strategies, fabrication, and applications. Read together they describe a field that has settled on a small handful of recurring motifs - nanopillar arrays, nanostar and nanorod ensembles, plasmonic-on-flexible-carrier composites, and ordered three-dimensional architectures - and is now competing on uniformity, shelf life, and unit cost rather than on raw enhancement factor.

The headline enhancement numbers have not moved meaningfully. The community still quotes electromagnetic enhancements in the 10^6 to 10^11 range depending on geometry and analyte, and single-molecule SERS continues to be a useful but not industrially relevant capability. The 2025 dynamic-SERS review in Nanoscale is explicit on this point: reproducibility and substrate uniformity remain the principal barriers to applying single-molecule sensitivity outside a research setting.

Flexible cellulose and tape-based substrates

The most active substrate family in the recent literature is paper- and tape-based. A 2025 review in Frontiers in Nanotechnology surveys cellulose-functionalised SERS sensors and is candid about the unresolved issue: cellulose roughness modulates the distribution of metal nanoparticles, which produces signal non-uniformity over the substrate area unless the deposition is carefully controlled. The reviewers describe several strategies - silver-nanorod glancing-angle deposition, silica-shell isolation of pre-formed nanoparticles, gold-on-tape inkjet routes - that each address one or two failure modes without yet converging on a single best practice.

For industrial users, the appeal is obvious: a cellulose or tape carrier is cheap, easy to handle, and lends itself to single-use workflows where contamination memory is the dominant risk. The flexible-substrate review in Coordination Chemistry Reviews notes the same point and identifies food-pesticide screening and environmental pollutant monitoring as the application areas where flexible SERS has the clearest pull. The 2026 Wiley review of Raman in dairy quality analysis reaches a compatible conclusion: SERS is positioned as a credible route to ultra-low detection limits for veterinary residues and trace contaminants, but the authors flag universal substrates with adaptive laser power control as a precondition for regulatory acceptance.

The translation to process analytics is partial. Flexible substrates suit at-line or grab-sample workflows much better than inline ones, because they are consumables and because their optical coupling is not designed for a fixed flow cell. Where SERS-in-bioprocess has gone inline, it has done so with rigid, microfabricated chips rather than with paper.

Three-dimensional architectures and additive routes

The 2026 PMC review on three-dimensional SERS substrates summarises the architectural side: ordered 3D scaffolds extend the hot-spot population from a two-dimensional area to a volume, which both raises analyte capture and increases tolerance to the angular variability that handheld or process geometries impose. The same review covers additive-manufacturing routes, including laser powder bed fusion of metal scaffolds and femtosecond-laser-printed nanopillar arrays, and treats them as a credible path to scalable, geometry-controlled substrates rather than as a curiosity.

The industrial relevance of additive-manufactured SERS is still emerging. The published demonstrations are characterised by reproducibility figures that are competitive with conventional lithography (relative standard deviations in the 6-15% range across a substrate area) and by cycle times that are short enough for batch production. What is missing is the supply-chain commitment: no major Raman vendor has yet announced a consumable line of 3D-printed SERS substrates with documented lot-to-lot performance and a standard part number. Until that exists, additive routes remain a research-grade option for groups that can fabricate in house.

The uniformity problem is the deployment problem

Across all of the 2025-2026 reviews, the common bottleneck is not enhancement, sensitivity, or chemistry - it is spatial and lot-to-lot uniformity. The Tian review puts this plainly: scalable fabrication and accurate control of nanogap geometry are the levers that move SERS from a single-paper demonstration to a deployable consumable. The Frontiers cellulose review reaches the same conclusion from the materials side, and the Nanoscale dynamic-SERS review from the optical-measurement side.

For an industrial buyer, this translates into a small set of due-diligence questions that the 2025-2026 literature now supports answering. What is the within-substrate RSD at the specified excitation, with what averaging area? What is the lot-to-lot RSD for a documented production batch? What is the storage stability under realistic conditions, and how is it characterised? Vendor data sheets that omit these numbers are not yet routine, but the academic literature has become specific enough that buyers can ask the right questions.

What this means for industrial use over the next year

Three plausible reads. First, flexible cellulose substrates will continue to dominate single-use and field-screening applications - food safety, environmental, and forensic - because their economics suit those workflows and the uniformity problem is bounded by the size of the interrogation spot. Second, three-dimensional and additive-manufactured substrates will begin to appear as vendor consumables for niche inline and at-line applications where geometry control is worth the unit cost. Third, the long-standing inline-SERS gap, where a flow-cell-mountable, calibration-stable SERS substrate would be commercially valuable, remains open. Nothing in the 2025-2026 literature signals an imminent product, although several research groups are clearly working on it.

For users tracking adjacent areas, the recent papers on time-gated Raman and on deep-learning chemometrics for Raman describe the measurement and modelling sides of the same problem set. SERS-substrate progress is one part of a wider effort to make Raman quantitatively credible at trace concentrations, and it makes more sense as part of that effort than on its own.