Optimized personnel flow with minimal contamination: development and validation of an air-barrier cleanroom for cell products processing

Abstract

Introduction

Cell processing facilities for regenerative medicine require strict prevention of cross-contamination. However, the typically employed sealed, multi-room layout increases energy demands and capital costs due to heating, ventilation, and air-conditioning (HVAC), and restricts staff mobility. We devised a semi-open (SO) cleanroom that eliminated doors between the cell processing room (CPR) and adjoining corridor, while maintaining unidirectional airflow as a barrier. This study rigorously compared four interface variants—plain opening, wing walls, push–pull ventilation, and a conventional swing door—to verify whether operational flexibility can be achieved without compromising particle content performance at the CPR–corridor interface.

Methods

Computational fluid dynamics (CFD) simulations reproduced two connected rooms separated by a 900 × 2000-mm² opening, supplied at 23 m³/min (35 air changes per hour) constantly. Four interfaces were evaluated: plain opening, 100–500 mm wing-wall panels, push–pull ventilation adjusted to a 0.75 ratio, and a conventional swing door. A 1-m/s cross-draft emulated personnel transit. Identical full-scale mock-ups were built; particle image velocimetry (PIV) quantified airflow vectors, and optical counters logged 0.5-μm aerosols during 5-min exit and entry. The primary endpoints were the inflow particle concentration ratio across the opening and the cumulative adjacent-room transfer proportions.

Results

CFD showed all layouts leaked ≤0.011 %, with a 1 m/s walking draft, push–pull kept inflow below 0.05 %, halving 500-mm wing-wall performance and outperforming plain openings. The PIV confirmed significant differences in airflow velocity distributions under each condition in the case of the exit. The semi-open layout without doors showed a lower proportion of vectors pointing opposite to the forward direction than the conventional layout in both the exit and entry cases. Particle counts supported this: push–pull transferred 0.013 % of particles on exit, 32.8 % on entry, giving overall migration to the adjacent room of 0.0043 %.

Conclusions

The SO cleanroom concept suppresses fluctuations in particle content at the CPR–corridor interface while eliminating physical doors, enabling flexible personnel flow and obviating extra HVAC zones. Push–pull ventilation delivered the most robust containment against walking-induced disturbances, whereas the 500-mm wing walls offered a passive, power-free alternative with moderate protection. With worst-case inter-room transfers below 0.05 %, SO designs can rationally replace conventional door-sealed rooms, substantially reducing energy and construction costs in regenerative medicine manufacturing.

Highlights

• •Semi-open cleanroom links rooms with an air curtain, eliminating physical doors.
• •Push–pull ventilation system restricts particle transfer to 0.013 % during operator exit.
• •A 10-fold lower corridor aerosol concentration than passive wing walls is achieved.
• •Simulation estimates that only 0.004 % of particles reach an adjacent room.
• •The proposed design reduces HVAC zones, slashing energy, and capital costs.

1. Introduction

Cell processing facilities (CPFs) manufacturing regenerative medicine products handle cells from different donors simultaneously, making rigorous cross-contamination control essential [[1], [2], [3], [4]]. Conventional CPFs mitigate contamination risk via process-specific zoning in classified cleanroom environments (Fig. 1) [5,6]. Maintaining such environments relies on high-efficiency particulate air (HEPA) filtration, and the associated heating, ventilation, and air-conditioning (HVAC) loads constitute a major component of operating costs [[7], [8], [9], [10]]. Therefore, the increasing energy burden has become a salient societal challenge in the pursuit of sustainable medical manufacturing systems. To further minimize cross-contamination, standard cleanrooms for cell product manufacturing segregate each processing step into separate rooms equipped with interlocking doors and pressure cascades. Although effective, this architectural approach increases capital expenditure. Moreover, physical isolation limits workforce mobility; once operators enter a room, they cannot readily move to another workspace, compelling single-task assignments and generating logistical and ergonomic stress during material transfers or rest breaks [11,12]. Consequently, flexible staff deployment is hindered, impeding workflow optimization and efficient labor allocation.

Fig. 1.

Schematic comparison of conventional and semi-open cleanroom layouts. The light blue arrows indicate the personnel flow.

Air-barrier SO cleanroom layouts that connect the cell processing room (CPR) and corridor (COR) with an air-curtain barrier have been implemented previously in Japan, as reported in the literature [[13], [14], [15]]. The SO concept eliminates rigid partitions and maintains a unidirectional airflow, thereby preventing particle migration to adjacent zones. Managing the CPR and COR as a single pressure zone eliminates the need for dedicated HVAC units, thereby reducing both capital expenditure and energy-related operating costs. In addition, the design improves worker mobility, enables flexible staff deployment, and expedites material logistics. Nevertheless, the potential risk of cross-contamination inherent in SO layouts has never been quantified, necessitating a systematic verification of their safety and effectiveness.

Four interface configurations between the CPR and COR were comparatively evaluated. These included: (i) an SO layout relying solely on supply-air flow to restrict particle transport, (ii) an SO layout augmented with wing walls flanking the opening to reinforce unidirectional flow, (iii) a push–pull ventilation system designed to actively block interzonal airflow, and (iv) a conventional door-sealed configuration. Effectiveness was assessed through computational fluid dynamics (CFD) simulations and performance testing in a full-scale mock-up cleanroom. Furthermore, assuming that assistant operators are shared across multiple production lines, we estimated the number of particles entrained into adjacent spaces during operator transit within the SO environment.

2. Materials and methods

2.1. CFD software and simulation parameters

Steady-state simulations were conducted using the thermal fluid solver, FlowDesigner 2024 (Advanced Knowledge Laboratory, Inc., Tokyo, Japan). The analysis model comprised two rooms, a CPR, and a COR, separated by an opening (900 mm (width) × 2000 mm (height)). To establish a unidirectional flow from the COR to the CPR unit, an air inlet (effective area 1250 × 1400 mm²; two 1250 × 700-mm² fan-filter units (FFUs) in tandem) was installed on the COR ceiling and an exhaust outlet (600 mm × 600 mm) was installed on the CPR wall. Both the supply and exhaust flow rates were set to 23 m³/min (35 air changes per hour), resulting in an average face velocity of 0.20 m/s across the opening. The room and supply air temperatures were maintained at 25 °C. The detailed boundary conditions are provided in Table 1.

Table 1.

CFD software and simulation parameters.

Software	Flow designer
Analysis area	5120 × 2800 × 2800 mm
Mesh division	69 × 38 × 38 = 99,636 mesh
Analysis target	Velocity, contamination
Analysis mode	Steady state analysis
Turbulence model	k-ε Turbulence model for high Reynolds number
Wall function	Logarithmic law + Spalding law

To assess containment at the opening, the CPR was designated as a contaminated zone and the COR as a clean zone. A particle source was positioned within the CPR and configured to conform to the European Union-good manufacturing practice (EU-GMP) Grade B operational limit, which specifies a maximum 0.5-μm particle concentration of 3.52 × 10⁵ particles/m³ under the current flow conditions. Accordingly, monodisperse 0.5-μm spheres (ρ = 1 g/cm³) were uniformly emitted into the CPR volume at 1.35 × 10⁵ particles/s.

2.2. Analysis condition assumed in CFD

The cases examined in this study are summarized in Fig. 2A. To ensure containment at the openings between rooms, three design variants were compared with the standard model (Fig. 2B and Table 2): (i) the installation of wing wall panels and (ii) the incorporation of a push–pull ventilation unit. The wing wall panels were designed to stabilize the unidirectional flow, and four projection lengths—0 (none), 100, 300, and 500 mm—were evaluated. The push–pull ventilation unit was modelled as two vertical panels positioned on either side of the doorway: a push panel delivering air into the opening and a pull panel extracting air from the opposite side. Based on prior findings demonstrating that a push–pull flow-rate ratio of 0.75 yielded the lowest airborne particle concentration at the opening [16], the push panel supplied 22.5 m³/min of air (panel-face velocity 0.37 m/s), while the pull panel removed 30 m³/min (0.49 m/s). This established the supply flow to 75 % of the extraction flow (22.5 m³/min ÷ 30 m³/min = 0.75). The resulting slight dominance of exhaust over supply produces a mild inward suction across the doorway, promoting unidirectional airflow toward the pull panel, thereby supporting particle containment within the CPR and on COR side.

Fig. 2.

CFD analysis for particle transfer to the adjacent room under baseline and personnel-movement conditions. (A) Schematic of the model cleanroom used in CFD analysis. Contamination is spread uniformly throughout the room in the form of 0.5-μm particles using the cell processing room. The lower figure shows the non-operating and operating conditions. During operation, human movement is simulated by reproducing a wind speed of 1.0 m/s using CFD. SA: supply air. EA: exhaust air. (B) Schematic of the installation of wing walls measuring 0, 100, 300, and 500 mm, and the push–pull ventilation system on the room side and COR side. (C) Particle distribution diagrams during non-operation (upper figure) and operation (lower figure) are shown in the Z section and Y section, respectively. The colors indicate particle concentrations. (D) IPC ratio calculated using CFD data. The left Y-axis shows values under non-operating conditions, and the right Y-axis shows values under operating conditions.

Table 2.

Experimental conditions for physical mock-up of the cleanroom.

Case	Not in operation	In operation
Inlet	Qin = 23 m3/min	Qin = 23 m3/min
Outlet	Qout = 23 m3/min	Qout = 23 m3/min
Particle generation	M = 135,000 [#/s]	M = 135,000 [#/s]
	Uniform generation in the cell processing room	Uniform generation in the cell processing room
Air-flow generation	–	V = 1 m/s
		A = 1700 × 370 mm at the room boundary

To approximate the transient disturbance caused by a person walking through the opening, a fan panel (1700 mm (height) × 370 mm (width)) was placed at the floor center and configured to emit a sweeping airflow of 1.0 m/s from the CPR toward the COR. The scenarios with a steady, undisturbed flow were classified as “not-in-operation,” whereas those incorporating the pedestrian-simulated jet were designated as “in-operation” (Fig. 2A).

CFD-predicted spatial fields of 0.5-μm particle concentration were volume-averaged over the COR to obtain the mean concentration C_COR [particles/m³]. A fixed source concentration of 3.52 × 10⁵ particles/m³ was imposed in the CPR. The inflow particle concentration (IPC) ratio was then calculated as a normalized measure of contaminant ingress for each design and operating scenario. The IPC ratio was defined as follows:

$$IPCratio=\frac{C_{COR}}{3.52\times 10^{5}paticles/m^3}\text{.}$$ (Eq. 1)

2.3. Construction of the full-scale mock-up cleanroom

A full-scale mock-up replicating the dimensions of the CFD model shown in Fig. 2 was constructed inside an existing ISO-class cleanroom (Fig. 3A–C). The original ceiling and grating floor were retained; however, new wall panels and doorways were constructed to reproduce the required geometries. The grating floor was sealed, and only the FFUs located above the test zone were operated. All other ceiling-mounted FFUs were deactivated to match the boundary conditions applied in the simulations (Fig. 3A). Four doorway configurations were prepared (Fig. 3B and C): (a) semi-open without wing walls, (b) semi-open with 500-mm wing wall panels, (c) semi-open with a push–pull ventilation unit, and (d) a simple swing door representing a conventional design. The airflow rates of the push–pull unit in configuration (c) were adjusted to ensure that the supply-to-exhaust ratio and face velocities matched those employed in the CFD analysis.

Fig. 3.

Structure of the mock-up cleanroom constructed for experiments. (A) Schematic of the mock-up cleanroom. R1, R2, C1, and C2 indicate the locations of particle counters, the asterisk marks indicate particle generation locations, the red arrows indicate human movement, and the gray arrows from the cell processing room to the next cell processing room indicate theoretical particle movement. As only one clean room can be installed for this verification, movement to adjacent rooms is assumed by verifying entry and exit in the same space. SA: supply air. EA: exhaust air. (B) Experimental conditions in the mock-up cleanroom. (C) Airflow visualization experiment using a laser light source. The green dotted lines visualize the microparticles generated by the laser light. The visualized laser light is recorded using a high-speed camera. The imaging conditions are kept the same for all conditions. (D) Actual images of the opening under each experimental condition. (E) Actual images under visualization conditions using a laser.

2.4. Particle image velocimetry (PIV) using mock-up cleanroom

Airflow was visualized using oil-mist tracer particles (diameter 1–100 μm) generated by a PORTA SMOKE PS-2005 generator (Dainichi Co., Ltd., Niigata, Japan) and illuminated by a sheet beam from a PIV Laser G6000 (Kato Koken Co., Ltd., Kanagawa, Japan) (Fig. 3D and E). Once the tracer behavior had stabilized, the illuminated field was recorded using a high-speed camera (Phantom Miro 110; Vision Research Inc., Charlottetown, Canada). The image sequence (1280 × 800 pixels) was analyzed via PIV (Flow-Expert v1.2.17, Kato Koken) using a 10-ms frame interval (100 fps), 0.1-m vector spacing and a 5-s analysis window.

For each experimental condition: (a) semi-open, no wing walls; (b) semi-open with 500 mm wing walls; (c) semi-open with a push–pull unit; and (d) swing-door (conventional), velocity vectors obtained at every measurement point were time-averaged over the first 5 s after the operator resumed walking following a 5 s pause. Vectors whose speed exceeded 0.5 m/s were treated as outliers and excluded. To assess whether the distributions of the rightward velocity magnitudes differed among the four conditions, pairwise two-sample Kolmogorov–Smirnov (KS) tests were performed using R v4.4.1. The x-direction component U (m/s) was exported, non-numeric and missing entries were removed with na.omit, and the ks.test (six unique pairs) was applied (Table 3, Table 4). As duplicate values were present, asymptotic p-values were used to adjust them using the Bonferroni method, with adjusted p < 0.05 considered as significant. Additionally, the number of rightward vectors directed from the CPR towards the COR within the region of interest (ROI) was counted and expressed as the proportion of COR-oriented airflow vectors.

Table 3.

Pairwise KS tests of airflow velocity distributions during personnel exit.

Pairwise comparison	KS D	Bonferroni-adjusted p
a_exit – b_exit ∗	0.106	7.3 × 10−11
a_exit – c_exit ∗	0.093	8.9 × 10−8
a_exit – d_exit ∗	0.129	1.0 × 10−16
b_exit – c_exit ∗	0.082	6.2 × 10−6
b_exit – d_exit ∗	0.197	9.6 × 10−39
c_exit – d_exit ∗	0.213	2.8 × 10−42

KS D is the maximum absolute distance between the two empirical cumulative distribution functions; larger D values indicate a greater divergence in the distribution shape.

Bonferroni-adjusted; α = 0.05. The p-values are approximate, owing to the ties in the data. ∗ Significant differences.

Table 4.

P-values for group comparisons of experimental airflow velocity during personnel entry.

Pairwise comparison	KS D	Bonferroni-adjusted p
a_entry – b_entry NS	0.031	1.00
a_entry – c_entry NS	0.031	1.00
a_entry – d_entry ∗	0.117	8.0 × 10−14
b_entry – c_entry NS	0.042	0.257
b_entry – d_entry ∗	0.120	3.0 × 10−14
c_entry – d_entry ∗	0.128	3.5 × 10−15

The KS D is the maximum absolute distance between the two empirical cumulative distribution functions; larger D values indicate a greater divergence in the distribution shape.

Bonferroni-adjusted; α = 0.05. The p-values are approximate, owing to the ties in the data. ∗ Significant differences. NS indicates no significant difference.

2.5. Particle-count measurements using mock-up cleanroom

The containment performance at the doorway was evaluated by measuring the airborne particle concentrations during the release of tracer particles (Fig. 3A). Two optical particle counters (Models 8306 and 8506-30, Particle Plus) sampled air every 10 s for 5 min; results for the 0.5–1.0 μm size channel are reported. The two movement scenarios for a single operator are defined as follows: Exit: the operator walked from the CPR into the COR; Entry: after crossing the COR, the operator entered the adjacent CPR. Both scenarios assumed a walking speed of 1.3 m/s and were initiated once the tracer levels stabilized. Under conditions (a–c), the operator paused for 5 s in the COR before proceeding, whereas under condition (d) (swing-door), the operator opened and closed the door for 5 s without an additional pause.

The adjacent-room particle transfer ratio was calculated as

$$\frac{\text{Particles}\text{in}\text{CPR}\left(\mathrm{R}1+\mathrm{R}2\right)}{\text{Paticles}\text{in}\text{COR}\left(\mathrm{C}1+\mathrm{C}2\right)}\text{,}$$ (Eq. 2)

where R1 and R2 are the cumulative counts inside the CPR, and C1 and C2 are the cumulative counts inside the COR, integrated over the entire 5-min sampling period. All concentrations are expressed in particles/m³.

2.6. Statistical analysis

Statistical analyses were performed using Prism 9 (GraphPad Inc., La Jolla, CA, USA) and R software. Data were presented as medians and interquartile ranges (IQR). Each statistical test is detailed in the corresponding figure legend. Statistical significance was defined as p < 0.05.

3. Results

3.1. CFD analysis for particle transfer to the adjacent room under baseline and personnel-movement conditions

The presence of the wing walls straightened the airflow, resulting in a stable unidirectional flow at the openings in both the horizontal and cross-sectional views (Fig. 2B and C). With the push–pull ventilation unit, the COR concentration remained low irrespective of the installation side. However, mounting the unit on the CPR side intensified internal recirculation and generated larger spatial gradients of velocity and particle concentration within the CPR (Fig. 2B and C). Under these undisturbed conditions, leakage from the CPR to the COR was negligible in all configurations: the highest IPC ratio was only 0.011 % for the doorway without wing walls (Fig. 2D).

The barrier jet at the doorway was weakened when a 1-m/s cross-draft—representing personnel movement—was introduced during operation, and the IPC ratio increased for both countermeasures, albeit to different extents. With the wing wall panels, increasing the projection length from 100 to 300 mm and then to 500 mm steadily lowered the particle concentration and the IPC ratio in the COR, confirming that an additional panel length enhanced leakage prevention (Fig. 2B and C). For the push–pull ventilation unit, the IPC ratios depended on the installation side; mounting the unit on the CPR side allowed more particles to escape into the COR, whereas installation on the COR side more effectively entrained and removed those particles (Fig. 2D). This is attributed to the fact that particles that leaked toward the COR side via the push-pull ventilation system were effectively removed when the system was installed on the COR side. However, this model assumes that the contamination source area encompasses the entire cell adjustment room and that the fan panel, which is an external disturbance source, is installed at the opening position, creating conditions where particle leakage is likely to occur. Therefore, the behavior of an actual cleanroom is expected to differ from that of this model. In this CFD simulation analysis, a comparison between the wing wall method and the push–pull ventilation system confirmed the superior barrier functionality of the latter against external disturbances under the given conditions.

3.2. PIV of airflow vectors during personnel exit

The moving average images of the airflow patterns, visualized using a high-speed camera and laser sheet, were analyzed under each layout (a: open, b: wing wall 500 mm, c: push-pull, and d: door) to observe the airflow conditions at the time of exit (Movies S1, S2, S3, and S4). For quantitative analysis, data were averaged over a 5-s period immediately after walking resumed following a 5-s pause, and the average vector was calculated (Fig. 4A). The X-axis component (velocity U m/s) was extracted from each image, and a frequency distribution plot was created for the range −0.3–0.3 m/s for comparison (Fig. 4B). The results showed that the airflow velocity distributions under each condition were significantly different among all six pairs (Bonferroni-corrected p < 0.05, KSD = 0.093–0.213; Table 4). The shape of the histograms showed significant overlap among the three improved conditions (a, b, and c), whereas condition d exhibited a peak shifted toward 0.05–0.20 m/s and a longer tail in the higher velocity range, indicating distinct characteristics compared to the other conditions. Furthermore, the proportion of rightward vectors pointing toward the COR (adjacent room) was calculated (Fig. 4C), with condition d showing the highest value of 64.2 %, followed by a (53.6 %), b (46.4 %), and c (45.7 %). In summary, the semi-open layouts without doors (a, b, and c) reduced the frequency of vector generation toward the COR compared to the conventional layout (d).

Fig. 4.

Experimental airflow velocity vectors during personnel exit. (A) Representative images for each condition. The strength of the airflow velocity vectors is indicated by color. Vectors with speeds exceeding 0.5 m/s are treated as outliers and excluded. (B) Frequency distribution of vectors at velocity U. This frequency distribution is statistically compared using the KS test. The results of the statistical analysis are presented in Table 3 (C) Proportion of vectors heading toward the CPR from the COR side.

3.3. PIV of airflow vectors during personnel entry

The airflow moving average of the entry conditions was visualized for each layout (Movies S5, S6, S7, and S8). For quantitative analysis, data were averaged over a 5-s period immediately after walking resumed following a 5-s pause, and the average vector was calculated (Fig. 5A). In the frequency distribution of velocity U within the range of −0.3–0.3 m/s, no significant changes were observed, as the room's air conditioning did not induce reverse flow (Fig. 5B). Comparisons between conditions showed significant differences in distribution only when compared with condition d (Bonferroni-corrected p < 0.05, KS D = 0.117–0.128). Therefore, similar airflow patterns were observed in conditions a, b, and c, whereas different airflow patterns were observed in condition d compared to those upon entering the room (Table 4). The proportion of vectors pointing away from the COR (adjacent room) was calculated (Fig. 5C), with condition d showing the highest value at 41.6 %, followed by a (32.1 %), b (32.7 %), and c (31.8 %). In summary, even under entry conditions, the semi-open layouts without doors (a, b, and c) reduced the frequency of vector generation toward the COR direction compared to the conventional layout (d).

Fig. 5.

Experimental airflow velocity vectors during personnel entry. (A) Representative images for each condition. The strength of the airflow velocity vectors is indicated by color. Vectors with speeds exceeding 0.5 m/s are treated as outliers and excluded. (B) Frequency distribution of vectors at velocity U. This frequency distribution is statistically compared using the KS test. The results of the statistical analysis are presented in Table 4 (C) Proportion of vectors heading toward the CPR from the COR side.

3.4. Airborne particle concentration dynamics during personnel exit

The 5-min trend of suspended particle counts (0.5 μm or larger) was observed during exit movements in each layout using artificially generated microparticles in the CPR (Fig. 6A). The arrows indicate the point at which the operator moves from the room toward the COR; the particle concentrations increased sharply from this point. R1 and R2 are particle counters installed inside the preparation room, whereas C1 and C2 are particle counters installed in the COR. The CPR maintained high particle counts due to the continuous artificial generation of particles, whereas in the COR, particle concentrations gradually decreased over time due to air conditioning (Fig. 6A). The total number of particles observed during the observation period was significantly higher at R1, located near the downstream exhaust port on the preparation-room side, under conditions c and d. In the COR, particle counts were higher under conditions b and d, and significantly lower under condition c (Fig. 6B). In other words, under condition c, very few particles migrated to the COR. Under condition d, despite the door being closed, particle concentration near the opening decreased after door opening, and shifted toward the COR side (Fig. 6B).

Fig. 6.

Airborne particle concentration dynamics during personnel exit. (A) Dynamics in the number of microparticles over 5 min. Data are collected six times and expressed as mean ± SD. The arrow marks the moment when personnel move from the CPR into the COR, highlighting the point at which particle-count dynamics begin to change. (B) Total particle number over 5 min at each measurement point. Data are presented as median values with IQRs and ∗ P < 0.05. The P-values are derived using the Kruskal–Wallis test with Dunn's multiple comparison test. (C) Adjacent-room particle transfer. The proportion of microparticles is believed to have moved from the cell processing room to the COR. Data are presented as median values with IQRs and ∗ P < 0.05. The P-values are derived using the Kruskal–Wallis test with Dunn's multiple comparison test.

The proportion of particles that migrated to the COR was calculated as the adjacent-room particle transfer (Fig. 6C). The results were a median of 0.176 % (IQR: 0.142–0.250). Under condition b, the value of median was 0.491 % (IQR: 0.351–0.665). Under condition c, the value of median was 0.013 % (IQR: 0.010–0.017). Under condition d, the median was 0.365 % (IQR: 0.265–0.495). These results indicate that the number of particles migrating to the adjacent room was extremely low under all conditions. In conclusion, the push–pull layout (condition c) reduced both the peak and total number of particles moving to the adjacent room upon exit, with a low migration of particles to the adjacent room.

3.5. Airborne particle concentration dynamics during personnel entry

The movement of artificially generated microparticles in the COR area was observed when entering the preparation room from the COR side during the entry process (Fig. 7A). As the air-conditioning system in the entire cleanroom created airflow from the COR side toward the preparation room, the microparticle concentration remained high regardless of operator entry, showing no significant changes (Fig. 7A). The concentrations of R1 and R2 in the CPR remained almost unchanged before and after entry; however, under conditions a and b, R1 and R2 showed higher levels than under other conditions (Fig. 7B). The microparticles generated on the COR side were highest under condition d, preventing their migration to the CPR in the adjacent room (Fig. 7B). Under condition c, the HEPA filters in the push–pull system reduced the particle concentration.

Fig. 7.

Airborne particle concentration dynamics during personnel entry. (A) Dynamics in the number of microparticles over 5 min. Data are collected six times and expressed as mean ± SD. The arrow marks the moment when personnel move from the COR into the CPR, highlighting the point at which particle-count dynamics begin to change. (B) Total particle number over 5 min at each measurement point. Data are presented as median values with IQRs and ∗ P < 0.05. The P-values are derived using the Kruskal–Wallis test with Dunn's multiple comparison test. (C) Adjacent-room particle transfer. The proportion of microparticles is believed to have moved from the COR to the cell processing room. Data are presented as median values with IQRs and ∗ P < 0.05. The P-values are derived using the Kruskal-Wallis test with Dunn's multiple comparison test.

The proportion ‵ particles that migrated to the adjacent room was calculated as the number of workers who exited (Fig. 7C). Under conditions a, b, c, and d, the median was 157 % (IQR: 140–193), 199 % (IQR: 172–232), 32.8 % (IQR: 30.0–36.8), and 50.0 % (IQR: 47.8–52.2), respectively. The values exceeding 100 % under conditions a and b were due to the absence of obstacles that impeded the airflow, necessary for maintaining the cleanroom, resulting in the accumulation of particles in the preparation room, with exhaust vents being counted. In summary, these results demonstrated that the push–pull layout under condition c reduced both the peak and total number of particles migrating to adjacent rooms upon entry.

3.6. Estimated number of particles moving to the adjacent room

The possibility of particles traveling from one preparation room, through a COR, to an adjacent preparation room where different manufacturing is being conducted is 32.8 % for 0.013 % of fine particles under condition c, even when preparation rooms are aligned. Therefore, 0.0043 % of the fine particles generated in the preparation room may migrate to an adjacent preparation room (Fig. S1).

4. Discussion

This study quantitatively benchmarked an air-barrier cleanroom as a “semi-open” design that eliminated physical partitions to address the practical challenge of balancing cross-contamination prevention and operational efficiency in a CPF. The containment performance of this SO layout was evaluated through CFD analysis and experimental verification in a mock-up cleanroom. Conventional fully separated structures using doors present challenges such as increased construction costs, high HVAC energy consumption, and inflexible personnel configurations. Furthermore, conventional door structures are not necessarily designed to prevent particle carryover, as opening doors inevitably induces airflow and particle movement. In contrast, the semi-open structure proposed in this study significantly suppressed particle migration between rooms while maintaining operator mobility. The push–pull method demonstrated the highest containment performance under both exit and entry conditions, suggesting its potential as an effective alternative for preventing cross-contamination. However, under all conditions, the amount of particulate matter carried into the adjacent rooms was extremely low, thereby reducing overall contamination risk.

CFD analysis is a powerful simulation method that enables non-invasive and quantitative prediction of airflow patterns and particle movement trajectories, making it highly effective for pre-evaluating containment performance during the design phase [[17], [18], [19]]. CFD analysis demonstrated that the installation of wing walls and the introduction of push–pull ventilation devices significantly reduced particle migration at openings. During “not in operation” conditions, particle migration rates remained extremely low, at 0.011 % or below across all systems. In contrast, under “in operation” conditions, particle migration increased due to external airflow disturbances; however, extending the wing-wall length tended to mitigate this effect in a stepwise manner. Although improvements can be achieved in the wing wall system by adjusting the wing-wall length, longer wing walls impose stricter layout constraints. Alternatively, establishing procedures and operational rules for worker movement between the cell adjustment room and the antechamber may further help suppress the frequency and intensity of external disturbances. Regarding countermeasure methods for openings, design considerations must balance particle leakage risks with the nature of products being handled and operational methods of the facility. Although the wing wall method offers advantages such as structural simplicity and no additional power requirements, its susceptibility to external disturbances remains a challenge. Additionally, when push–pull ventilation devices were installed on the COR side, the particle concentrations in the COR side further decreased, suggesting that the installation location significantly influenced performance. In summary, the wing wall method is structurally simple yet effective, whereas the push–pull method provides the most robust containment against external disturbances.

PIV is an experimental technique that enables high-temporal and high-spatial resolution visualization of airflow and quantification of velocity vectors. It is an excellent method for quantitatively evaluating temporary airflow fluctuations associated with human movement [20,21]. In the experimental evaluation of airflow vectors using PIV analysis, compared to conventional methods, the frequency of airflow vectors toward the adjacent room was suppressed under semi-open conditions. During exit conditions, the conventional door condition exhibited the highest COR-directed vector ratio at 64.2 %, whereas the semi-open layout condition remained within the range of 45 %–54 %. Similarly, during entry, the door condition showed the highest percentage with statistically significant differences compared to the other conditions. This is attributed to the generation of airflow that follows the pedestrians' backs in response to changes in airflow when the door is opened or the airflow generated by the door itself. Previous reports have also reported that the turbulent vortices generated when doors are opened cancel out the pressure differences and cause air exchange between clean areas [22]. Therefore, airlock rooms are considered ideal. However, some facilities are unable to install airlock rooms due to architectural cost constraints or limitations in available floor space, necessitating operational precautions. In contrast, semi-open layouts maintain a constant unidirectional airflow, which reduces the likelihood of transient reverse flows. The results of this study also demonstrated that in doorless layouts, the frequency of airflow toward adjacent rooms was suppressed, and the presence of an “invisible wall” of air effectively separated clean areas, consistent with previous reports [[23], [24], [25]]. These results suggest that temporary pressure and airflow changes caused by door opening and closing are primary factors in particle migration to adjacent rooms, thereby supporting the effectiveness of unidirectional airflow control in open structures. However, in spaces lacking complete physical barriers, potential risks such as contamination due to human error must be considered. Accordingly, operational procedures and essential training must be established and implemented in tandem to ensure appropriate management.

Particulate matter measurement enables real-time, direct assessment of cross-contamination risks and is therefore critically important as an evaluation metric for containment performance. Particles with a diameter of 0.5 μm or larger are specific indicators under EU-GMP standards, and their monitoring is essential for ensuring cleanroom performance [7,26,27]. In the analysis of particle number dynamics, the push–pull ventilation system minimized particle migration to adjacent rooms compared with the other systems, under both exit and entry conditions. Particularly during exit, the push–pull ventilation conditions resulted in an extremely low particle migration rate to the COR, and even under other conditions, the migration rate remained below 1 %. These measured data demonstrate a cross-contamination risk that is orders of magnitude lower than previously reported particle leakage during door opening. For example, approximately 5 % air leakage has been reported in negative-pressure isolation rooms, and door operations in positive-pressure operating rooms have been shown to cause air exchange of several tens of percent [22]. Additionally, theoretical calculations of particle movement from the preparation room to the COR and further to the adjacent room under condition c yielded a total migration rate of 0.0043 %, an extremely low value. In practice, even such a minimal number of particles migrating to an adjacent room is unlikely to penetrate air curtains or devices such as safety cabinets. Furthermore, based on reports indicating that 0.08 % of particles in a clean environment are contaminated with microorganisms [26], 0.0034 bacteria can migrate to the adjacent rooms. Airborne bacteria vary depending on the environment and individual workers; therefore, this value may not always be applicable, but can serve as a reference. Based on the aforementioned evidence, the proposed semi-open layout can be concluded to have a practical utility in ensuring safety during the manufacturing of regenerative medical products. This suggests that the semi-open layout proposed in this study can serve as an optimal solution that balances operational efficiency and containment performance.

The first limitation of this study is that it was restricted to physical experiments using a mockup facility, and performance evaluations under actual environmental conditions, such as temperature and humidity fluctuations, long-term operation, and compliance with EU-GMP requirements, were not conducted. For example, even a slight change in the position of the supply air from a straight front-facing flow to an angled one can alter the airflow, potentially increasing or decreasing the risk of particles reaching the adjustment room. In future studies, long-term operational tests in various facilities and particle movement analyses during actual cell manufacturing processes will be necessary. The second limitation is the discrepancy between the results of the CFD verification and those of the mock-up cleanroom experiments. Compared to CFD, the detection risk in the mockup cleanroom was significantly lower. This is likely due to the CFD simulation being conducted under single-sided conditions without incorporating door operations, making direct comparisons challenging. However, the CFD verification was considered conservative, which possibly resulted in the maximum particle levels being accumulated on the adjustment room side. Despite this, the results from the prior CFD and the mock-up cleanroom were generally similar, supporting the utility of CFD as a tool for assessing the relative magnitude of contamination risks.

This study demonstrated that a semi-open structure—particularly a push–pull ventilation system—has the potential to significantly reduce cross-contamination risks while improving operational flexibility, serving as an effective alternative to conventional door-equipped cleanrooms, as demonstrated through CFD analysis and actual facility experiments. Although the wing wall-equipped structure demonstrated similar effectiveness, the push–pull system outperformed it under conditions involving external disturbances. The adoption of semi-open cleanrooms is expected to alleviate the stress associated with personnel movement and material handling, thereby enhancing the overall efficiency of CPF operations.

Authors' contributions

MM, HT and KN; Data acquisition, analysis, and interpretation: MM; Manuscript drafting: MM, KA, HT, GT, DS, YC, KN, and IS; Manuscript revision to address important intellectual content. All the authors have read and approved the final manuscript.

Funding

This research was funded by a joint research grant from the Takenaka Corporation.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests:Mitsuru Mizuno reports financial support was provided by Takenaka Corporation.

Appendix A. Supplementary data

The following are the Supplementary data to this article.

Movie S1

Movies of particle image velocimetry (PIV) analysis under condition a for exit. Representative airflow visualization in a cleanroom environment. All movies (S1–S8) depict 10-s sequences acquired using particle image velocimetry (PIV). For analysis, velocity vectors were time-averaged over the first 5 s after the operator resumed walking following a 5-s pause. Vectors exceeding 0.5 m/s were excluded as outliers. This processing is consistent across all supplementary movies.

Movie S2

Movies of PIV analysis under condition b for exit.

Movie S3

Movies of PIV analysis under condition c for exit.

Movie S4

Movies of PIV analysis under condition d for exit.

Movie S5

Movies of PIV analysis under condition a for entry.

Movie S6

Movies of PIV analysis under condition b for entry.

Movie S7

Movies of PIV analysis under condition c for entry.

Movie S8

Movies of PIV analysis under condition d for entry.