---
title: Open-Source MPWs for CMOS Biosensors
url: https://www.emergentmind.com/topics/open-source-multi-project-wafers-mpw
type: topic
---

# Open-Source MPWs for CMOS Biosensors

Searching arXiv for the cited MPW papers and related terminology to ground the article in current literature.
Open-source multi-project wafers (MPWs) are a specific MPW model in which multiple participating groups submit designs into a shared foundry run and, crucially, all designs on the wafer are open-source and visible to all participants. In the implementation described for CMOS bioelectronic sensor prototyping, the program is the Nanotechnology Xccelerator, a “pilot multi-facility MPW program that employs open-source tools for generating CMOS designs.” Unlike conventional MPWs, where participants largely remain siloed and only receive their own dice, this model lets participants receive either individual flash fields containing all projects or entire wafers, while also giving everyone access to the designs submitted by others. Within the reported biosensor workflow, open-source MPWs function not only as a cost-sharing fabrication mechanism but as enabling infrastructure for shared design reuse, wafer-scale back-end-of-line (BEOL) processing, and modular heterogeneous integration [2509.24075].

## 1. Definition and collaboration model

In the reported formulation, an open-source MPW is distinguished from a conventional MPW by the visibility and reuse of layouts across the participant cohort. The defining feature is not merely that multiple projects share a wafer, but that every design on the wafer is open-source and accessible to all participants. The paper states that each participant can receive either individual flash fields containing all projects or entire wafers at a low cost to allow optimum post-processing for the final application, rather than only isolated project-specific chips [2509.24075].

This changes the collaboration model from isolated IP-preserving participation to shared design reuse across projects. The platform explicitly encourages participants to inspect and reuse other participants’ structures and to support collective development of measurement test structures. In the authors’ framing, this matters particularly for CMOS bioelectronic sensors, where commercialization and prototyping are hindered by custom silicon access, expensive mask sets and EDA tools, repeated design/fabrication/testing loops, packaging and heterogeneous integration, and heavy post-processing for fluids and interfaces.

A common misconception is that an open-source MPW is simply a lower-cost version of a conventional MPW. The reported distinction is broader: design access, delivery format, and post-processing strategy all change. This suggests that the open-source MPW is best understood as a shared design ecosystem rather than only a fabrication service.

## 2. Wafer-level prototyping and BEOL manufacturability

The wafer-scale workflow is organized around a BEOL-oriented prototyping strategy. Foundry fabrication stops after the last set of inter-layer vias and planarization, so the returned wafers are planar and unpassivated, with surface roughness of about \( \approx 1 \,\text{nm} \) and maximum step height of about \( \approx 10 \,\text{nm} \). This mid-process wafer then becomes the substrate for BEOL post-processing, with all steps kept at or below \(300\,^{\circ}\text{C}\) because the wafer must not exceed the thermal budget that would damage CMOS [2509.24075].

The BEOL sequence adds both protection and routing. Ti pads protect vias and act as etch stops. Low-stress silicon nitride passivates and blocks mobile ions. Windows are opened to access selected regions. Ti backfill restores electrical access. A fan-out layer routes signals to large peripheral pads. Final electrode metals are added for the working and reference electrodes. Wafers are diced only after BEOL completion, and chips are then assembled into modular fluidic and electrical packages.

The authors explicitly state that this MPW strategy improves manufacturability, process yield, reuse of instrumentation and infrastructure, and the ability to create standardized interfaces for multiple projects. A key practical point is that the same redistribution strategy can be reused across wafer projects, which promotes a common electrical interface. Because the returned wafer is planar and unpassivated, it is more amenable to standardized BEOL post-processing than fully packaged small dice.

The fan-out geometry is a central yield-related feature. Leads are routed over unused projects in the reticle, producing a large handle area of approximately \(31\,\text{mm} \times 25\,\text{mm}\), while nitride isolation helps electrically isolate unused projects and minimize failures. In the reported interpretation, this is a practical yield enhancement because it creates a larger, easier-to-handle substrate than the original small chips and supports higher-throughput lithography.

## 3. Modular heterogeneous integration

The reported platform treats heterogeneous integration as a modular packaging problem organized into electrical, mechanical, and fluidic layers. The mechanical interface is provided by a base support that holds the chip, provides positive registration, constrains alignment with guide features, and supports repeatable chip insertion and removal. It can be fabricated by CNC machining or 3D printing, and the authors note that it could also be adapted for injection molding [2509.24075].

Electrical access is handled by the BEOL redistribution layer and large peripheral pads of about \( \approx 1\,\text{mm} \) per edge with pad pitch of about \( \approx 1.7\,\text{mm} \). Contact is made through spring-loaded pogo pins on a PCB. The large pad geometry makes the system tolerant to angular alignment errors of up to about \( \pm 1^\circ \). The design eliminates wire bonding and soldering, enabling rapid chip exchange. A loopback LED scheme is used to validate contact and alignment: when the pogo pins correctly contact the pads, the LEDs illuminate, confirming registration and orientation.

The fluidic interface is implemented through modular inserts threaded into the top of the package. In the described work, the authors used an open well insert with volume \( \approx 160\,\mu\text{L} \), suitable for pipetting and rapid exchange. They note that this insert can be swapped for other designs, including microfluidic-channel inserts for faster exchange.

The significance of this layered packaging strategy lies in reuse and decoupling. The chip, redistribution, electrical contacting scheme, and fluidic module are not treated as a single monolithic package. Instead, the mechanical structure and interface stack are modular and reusable, while the sensor chip may be application-specific.

## 4. Sensor-array demonstration and analytical framework

The biosensor demonstration uses a 25-unit-cell amplifier array designed to measure DNA hybridization. The system includes 25 independently addressed working electrodes per chip, with 10 chips providing a total of 250 electrodes for analyte measurements and 3 control chips providing a total of 75 control sensors. The reported average capacitance change upon hybridization is \( (24.5 \pm 3.9)\,\text{pF} \), corresponding to a relative change of \( (7.8 \pm 0.9)\% \), with \( k = 2 \), \( \approx 95\% \) confidence. For the control using a non-complementary strand, the average change is \( (-5.5 \pm 4.8)\,\text{pF} \), corresponding to \( (-1.3 \pm 0.9)\% \), also with \( k = 2 \), \( \approx 95\% \) confidence [2509.24075].

The front end is modeled using a Randles-circuit-based impedance model, and the intended operating regime is one in which the amplifier gain is proportional to the sensing capacitance, with \( G = -\frac{C_{DNA}}{C_F} \). The stated design choices include \( R_F = 2~M\Omega \), \( C_F = 10~pF \), \( R_{SOL} = 1~k\Omega \), and \( C_{DNA} = 10~pF \) or \(100~pF \). The amplifier’s practical operating window is reported as 10 kHz to 100 kHz, with a lower cutoff of \( f_L = \omega_L/2\pi = 7.95~Hz \), and experimentally good agreement below 100 kHz.

For classification, the feature vector is defined by
\[
x_1 = C_i
\]
and
\[
x_2 = \frac{C_f - C_i}{C_i},
\]
where \( C_i \) is initial capacitance and \( C_f \) is post-exposure capacitance. The relative likelihood is defined as
\[
R(x)=N(x)/P(x),
\]
and the probability that a sensor detects analyte is
\[
\Pr[+|r] = \frac{q}{q+(1-q)R(x)}.
\]
The paper notes that the \( R(x)=1 \) boundary can be used as a classifier if prevalence \( q \) is unknown, and that for the presented data the characteristic accuracy is at least 95% for any prevalence.

In this setting, the machine-learning analysis is tied directly to MPW-enabled sensor parallelism. The array data are inherently parallel and variable across sensors, and the classification framework turns that parallel structure into a statistically interpretable diagnostic platform. This allows the assignment of individual sensors to positive or control classes with an explicit probabilistic interpretation.

## 5. Relationship to conventional and emerging MPW models

The open-source MPW reported for CMOS bioelectronic sensors is one point within a broader MPW landscape. Conventional commercial MPW services remain important because they lower the cost and iteration time of prototyping by allowing a user to purchase a portion of a wafer rather than fund a dedicated run. In silicon nitride photonics for astronomy, MPW platforms from foundries such as Ligentec and LioniX are presented as a practical route for exploring device architectures and trade-offs before committing to dedicated fabrication. That work estimates that an AWG with \( R \sim 10{,}000 \) can be developed with MPW runs and that, depending on fabrication and design optimizations, throughputs \( \sim 60\% \) are possible on the SiN platform [2106.04598].

The distinction is that those commercial MPW platforms are described primarily as affordable access to mature fabrication processes, whereas the open-source MPW for biosensors modifies design visibility, delivery format, and BEOL reuse. In the biosensor setting, participants receive flash fields containing all projects or entire wafers, and all submitted designs are visible to all participants. The result is a shift from physical co-placement alone to shared design reuse and standardized interface development.

A further related, but distinct, evolution is active on-die multi-project integration. The Chipstitch work argues that traditional MPW services based solely on physical co-placement do not scale when many small projects each need pads, memory, control logic, and power-measurement support. It therefore proposes active chip-site aggregation with algorithmic packing, narrow-area shared interconnect, and perimeter-based power shutdown domains. Experimental results show area reductions of up to 13x over state-of-the-art physical-only aggregation methods [2512.10089].

These three models—conventional MPW, open-source MPW, and active on-die aggregation—address related constraints but are not interchangeable.

| Model | Defining feature | Reported emphasis |
|---|---|---|
| Conventional MPW | Users buy a portion of a wafer | Lower cost and iteration time |
| Open-source MPW | All wafer designs are open-source and visible to all participants | Shared design reuse, BEOL post-processing, standardized interfaces |
| Active on-die aggregation | Shared on-die infrastructure replaces physical-only co-placement | Scalable support for many small independent designs |

This comparison clarifies another common misconception: open-source MPW is not synonymous with any multi-project sharing scheme. In the reported literature, it names a specific access and reuse model, while active aggregation refers to a different architectural response to scaling pressure.

## 6. Significance, limitations, and likely implications

Within the reported biosensor program, the open-source MPW is not presented as a manufacturing convenience alone. It is described as the enabling infrastructure for wafer-scale biosensor prototyping because it lowers cost, shares design knowledge across participants, enables reusable BEOL post-processing, improves manufacturability and yield, supports standard electrical and fluidic packaging, and makes large parallel sensor arrays practical for biological assays [2509.24075].

Its practical significance is especially clear in relation to the usual bottlenecks of CMOS bioelectronics. Conventional MPW chips are described as small and hard to handle, and the usual packaging steps can be low-yield and labor-intensive. By returning planar, unpassivated wafers at a stage suitable for low-temperature BEOL fabrication, the reported flow allows exposed electrodes, signal redistribution, ion-blocking passivation, and customized surface chemistry to be added in a standardized manner. This is why the authors state that the approach supports advanced BEOL fabrication techniques and improves the manufacturability and process yield of CMOS biosensors.

The approach also has clear boundaries. The sensor designs themselves may be single-use, and the paper does not claim that every element of the system is universally reusable. Rather, the reusable components are the tooling and interface stack: redistribution, pad geometry, package mechanics, electrical contacting, and interchangeable fluidic modules. That distinction is important for interpreting the scope of “reuse” in open-source MPWs.

A plausible implication is that such a platform can shorten repeated design/fabrication/testing loops by combining shared layouts, standardized post-processing, and modular interfaces in a single wafer-scale pipeline. The broader strategy, as described, is to move CMOS biosensors from isolated, hard-to-package prototypes toward a repeatable wafer-scale development platform in which the chip, post-processing, packaging, data acquisition, and analysis all fit into a manufacturable and reusable pipeline.

Source: https://www.emergentmind.com/topics/open-source-multi-project-wafers-mpw