Trends and Challenges in Multistep Continuous Flow Synthesis

Abstract

Flow chemistry has become an important tool for manufacturing nowadays, driven by the growing demand for safety, efficiency, scale, and sustainability in chemical industries. Building on basic flow chemistry techniques, multistep continuous flow synthesis has emerged as an attractive synthetic process in pharmaceutical, agricultural, and materials synthesis. In practice, numerous issues need to be addressed in multistep continuous flow processes, such as reaction compatibility, step-connection, material transfer, and flow stability. Aiming to establish an industrially viable multistep flow process for tiadinil, methiadinil, emamectin benzoate and other drug molecules in agricultural and pharmaceutical industry, we cooperated with enterprises in the past few years and encountered nearly all the difficulties that may arise in the process, and these difficulties are also long-standing challenges for industry settings. To tackle these barriers faced, we developed and searched for various solutions to address them. These practices further facilitated the formulation of the common solutions for developing a multistep continuous flow process. This perspective will integrate relevant literature with our research to showcase the key challenges in multistep continuous flow synthesis and discuss corresponding solutions. Additionally, general development of flow process usually begins with extensive screenings of reaction conditions in batch before further optimization in flow. Instead, a more efficient approach called de novo flow, defined as performing reaction optimizations primarily in flow after feasibility has been established in batch, has been emerging as a viable strategy. We also present an outlook on the forthcoming development of multistep continuous flow synthesis and point out the key issues to be addressed in the future.

1. Introduction

Flow chemistry, in which reagents are continuously fed through a reactor, has facilitated continuous manufacturing in selected applications and can provide an alternative to conventional batch processing. By maintaining a steady flow through tubular or microchannel reactors, it enables precise control over reaction parameters such as temperature, pressure, and residence time. This leads to enhanced mass and heat transfer, selectivity, reproducibility, and safety , while scaling production can also be achieved via parallelization (numbering up) rather than simply increasing the vessel volume. Continuous flow technology is also particularly advantageous when dealing with hazardous or highly reactive intermediates that are sensitive to temperature or mixing. −

Continuous flow reactions have so far become the primary production method in large-scale chemical and petrochemical industries. In industrial manufacturing sectors such as pharmaceuticals and pesticides, batch reactor operations still dominate. The application of microreactors-based continuous flow reactions is even more delayed. In fact, as early as 1959, Richard Feynman suggested miniaturization as the future direction of scientific development. In the 1970s, Ramshaw from Imperial Chemical Industries in the UK first proposed the concept of microreactors. A highly significant milestone was the development of a micro heat exchanger in 1989 at the Karlsruhe Research Center in Germany. Subsequently, companies such as DuPont, Bayer, and BASF conducted ongoing research on microreactors and widely adopted them for industrial production. ,

Among the many applications of flow chemistry, single-step continuous flow reactions have become a cornerstone, offering improvements in reaction rates, safety, and reproducibility over conventional batch processes. These reactions, ranging from photochemical and microwave reactions to electrochemical reactions, leverage the intensified heat and mass transfer within the flow system, enabling faster and more efficient syntheses. Photochemical flow reactions, for example, utilize short optical path length in compact photoreactors to improve reaction efficiency and scale, overcoming partial limitations of batch photochemistry. Electrochemical flow reactions take advantage of smaller interelectrode gaps to reduce resistance and enhance the efficiency of redox processes compared with batch protocols, further pushing the boundaries of what can be achieved in continuous flow systems. −

Electrophotocatalytic (EPC) has emerged as a powerful strategy in organic synthesis by integrating the advantages of both photochemistry and electrochemistry. EPC can overcome some limitations of traditional methods and its selectivity and reaction efficiency can be enhanced by flow chemistry. For instance, Reek, Noël and co-workers have made significant contributions to flow-based EPC by developing reactors that combine both photochemical and electrochemical processes in a single system. , They have successfully demonstrated this in reactions such as C­(sp³)-H heteroarylation, where the use of a flow reactor enhanced both reaction rate and selectivity, offering a scalable method for the late-stage functionalization of drug molecules.

The efficiency of process development for microflow reactions has been unsatisfactory over the years, which in turn has constrained their broader application and widespread adoption. In terms of research and development (R&D) methodology, the conventional approach still involves screening reaction conditions in batch mode before switching to continuous flow. Data obtained from batch reactions often does not match continuous flow systems, leading to considerable waste of time and resources. In current industrial production, microflow synthesis is usually used in only one or two reaction steps. The synthesis of functional molecules, in contrast, always involves multiple chemical transformations, and multistep continuous reactions are the foundation of future smart factories and dark plants. It is anticipated that the new generation microflow technology will overcome these challenges by featuring multistep continuous or end-to-end synthetic processes, high degree of automation and intelligence, and a new research paradigm known as de novo flow.

Compared to single-step flow reactions, multistep ones face significantly more complex challenges. − In comparison with single-step reactions, multistep processes require compatibility between consecutive reactions, integration of in-line workup steps, and management of issues such as solids formation, phase separation, and solvent incompatibilities. Key challenges that need to be addressed include maintaining reaction compatibility, flow stability, and optimizing scale-up strategies, etc.

Textbooks, − reviews − and perspectives , have presented significant progress in developing strategies and technologies to facilitate multistep flow synthesis. It is still relatively insufficient to take the common key challenges encountered in multistep continuous flow synthesis as the main discussion framework for reviews or perspectives, and the solutions to such challenges are infrequently discussed systematically. In this perspective, we will provide a detailed discussion of the key challenges in multistep continuous flow synthesis and discuss possible solutions. It will also offer an outlook on the future of multistep continuous flow synthesis, offering insights into its ongoing development. The legend for flow components is presented in Figure .

1.

Legend for flow components.

2. Challenges in Multistep Continuous Flow Synthesis

In our recent work on the multistep continuous flow synthesis of tiadinil and methiadinil (Figure ) and emamectin benzoate, we have encountered nearly all the typical challenges that may arise in multistep continuous flow synthesis. We found that solutions to these challenges are often scattered across numerous studies, but comprehensive reviews or perspectives that systematically address both the problems and solutions specific to multistep continuous flow synthesis remain relatively scarce. And what mainly influences the reactivity in continuous flow chemistry are factors such as mass and heat transfer efficiency, solvent selection, reactant concentration and reactor geometry and size. It also includes the specific reaction conditions and corresponding equipment: gas–liquid reactions (Figure , Step 2), solid–liquid reactions (Figure , Step 1), reaction compatibility matching (Figure , Step 4), in-line monitoring, in-line workup (Figure , Step 3), scale-up issues, de novo flow and AI-assisted flow synthesis.

2.

Multistep continuous flow synthesis of tiadinil and methiadinil by He, Gao and co-workers. DCE: 1,2-dichloroethane. HOBt: 1-hydroxybenzotriazole. DMF: N,N-dimethylformamide. EDCI: N-(3-(dimethylamino)­propyl)-N′-ethylcarbodiimide hydrochloride.

The following sections will focus on a detailed discussion of the challenges in multistep continuous flow synthesis, combining both our own research experience and the solutions proposed in the literature.

2.1. Gas–Liquid Reactions

In addition to these issues, gas–liquid reactions in continuous flow still face several fundamental challenges to be solved.

2.1.1. Challenges in Gas–Liquid Reactions

The core issue of gas–liquid reaction lies in its extremely low solubility, which in turn results in low mass transfer efficiency. For example, many synthetically valuable gases such as CO, CO₂, and even O₂ exhibit low solubility in common organic solvents, requiring elevated pressures to reach useful concentrations. And the achievable interfacial area and diffusion length are highly dependent on reactor type and flow regime, making it difficult to maintain consistent gas uptake across scales.

It is because of the low solubility of gases that flow-regime sensitivity arises, which subsequently gives rise to operational uncertainty. Gas–liquid multiphase flow easily transitions between bubbly flow, Taylor flow, Taylor-annular flow, annular flow or misty flow regimes, depending on viscosity, channel dimension, and relative flow rates. Each regime possesses markedly different interfacial areas and residence-time distributions, influencing both selectivity and conversion in unpredictable ways.

Gas–liquid reactions in both laboratory and industry processes may encounter a series of problems depending on the specific conditions, which can be categorized into gas-fed reactions and gas-generating reactions.

2.1.2. Gas-Fed Continuous Flow Reactions

Gas-fed reactions refer to processes where gases are directly introduced into the continuous flow system. The primary concern of gas-fed reactions is to ensure precise dosing and uniform distribution of gases throughout the reaction system. Inaccurate gas delivery can lead to inconsistent reaction rates, affecting both selectivity and conversion. Maintaining a stable gas flow under high-pressure conditions is critical because most gases may require elevated pressures for efficient dissolution in the liquid phase. Regarding gas delivery issues, various resolution strategies are accessible for consideration.

2.1.2.1. Back-Pressure Regulator in Gas-Fed Reactions

Typically, the most common solution to these challenges is the use of a BPR. By integrating a BPR, the entire reactor network can be uniformly pressurized, thereby ensuring stable gas delivery. This can enable reactions to occur above the normal boiling points of solvents and enhance the solubility of gaseous reagents, which is essential for an efficient mass transfer rate. Additionally, continuous flow systems commonly do not experience the issue of high gas concentrations accumulating in the headspace, which is often seen in batch equipment and it can make gas stoichiometry more predictable via pressure adjustments. These characteristics make flow systems especially effective for handling immiscible phases and accelerating gas-involved chemistry that would otherwise be constrained by slow mass transfer or safety concerns.

Zhu et al. developed a fully continuous flow synthesis of 2′-deoxy-2′-fluoro-arabinoside (FAU, 2–14), a key intermediate for the antiviral drug azvudine, employing a six-step continuous flow process (Figure ). Throughout the process, a BPR was integrated to uniformly pressurize the reactor network, thereby enhancing the solubility of hydrogen chloride gas and improving its mass transfer efficiency. Utilizing mass flow controllers (MFCs) enables precise delivery of gas. Under a back pressure of 75 psi, at a temperature of 50 °C, and with a residence time of 30 min, the chlorination reaction for the conversion of 2–8 to 2–9 was successfully accomplished. Subsequently, following the processes of hydrolysis, fluorination, and bromination, sodium carbonate was employed as a quench agent to neutralize the acid added in the reaction, along with the releasing of carbon dioxide gas.

3.

Fully continuous flow synthesis of 2′-deoxy-2′-fluoro-arabinoside (2–14) by Zhu et al. DCM: dichloromethane. DAST: diethylaminosulfur trifluoride. MeOH: methanol.

Thanks to the presence of the BPR, the reaction solution was able to pass smoothly through the liquid–liquid separator and ultimately flow into the surge vessel for drying. Finally, through the processes of condensation and deprotection, FAU was successfully synthesized via a six-step continuous flow process with an overall yield of 32.3% and a residence time of 156 min.

2.1.2.2. Ultrasound in Gas-Fed Reactions

Most multistep continuous flow reactions involving continuous gas introduction can be addressed with a BPR. In some cases, only using a BPR may not be sufficient to ensure that the gas–liquid reaction proceeds completely. Under such circumstances, replacing the reactor can also be considered as one of the alternative solutions. Various designs of gas–liquid microreactors have been developed to enable gas–liquid reactions. Chen et al. developed a high-power ultrasonic microreactor (USMR) to enhance gas–liquid mass transfer in microfluidic systems. The USMR combines a microreactor plate with a Langevin-type ultrasonic transducer, functioning as a longitudinal half-wavelength resonator to create a uniform acoustic field. Its design includes a piezoelectric ceramic transducer between front and back masses, optimizing energy efficiency. The USMR improves mass transfer by 3.3 to 5.7 times through noninertial cavitation on slug bubbles, driven by stable surface wave oscillations and microstreaming. This makes it an effective tool for enhancing mixing in gas–liquid reactions, leading to improved gas solubility and greater stability of the flow regime.

2.1.2.3. Spinning Disk Reactor in Gas-Fed Reactions

The spinning disk reactor (SDR) has emerged in recent years. , It is used in gas–liquid reactions and typically operates under high gravity conditions, facilitated by centrifugal forces generated by the high-speed rotation of the reactor’s packed bed. − In the SDR, liquid and gas phases interact in a countercurrent flow pattern. The liquid is distributed in the form of films, droplets, and liquid ligaments, which create a large surface area for effective mass transfer between the phases. This high surface area, combined with the rapid mixing capabilities of the rotating packed bed, enhances the efficiency of processes like absorption, distillation and chemical reactions that involve gas–liquid interactions. And the reactor’s structure, often designed with mesh or wire packings, is optimized to provide enhanced interphase contact and to overcome some limitations of traditional reactors, such as slower mass transfer rates. SDR is suitable for both laboratory and industrial applications, offering high performance in a compact design.

2.1.2.4. Tube-In-Tube Reactor

A collaborative effort among O’Brien, Ley, Polyzos and co-workers resulted in the development of a tube-in-tube reactor device. It consists of two concentric capillaries, with a gas-permeable membrane made from Teflon AF-2400 in the inner tube. Due to the unique properties of Teflon AF-2400, which allows high gas permeability while remaining impermeable to liquids, the reactive gas is introduced into the inner tube and reacted with liquid phase. This reactor’s ability to use only small volumes of pressurized gas makes it ideal for high-pressure reactions. Also, the system’s scalability and flexibility make it suitable for applications in key C–C, C–N, and C–O bond forming and hydrogenation reactions. And the integration of real-time monitoring systems allows for continuous optimization of reaction conditions, enhancing both safety and efficiency.

2.1.3. Gas-Generating Continuous Flow Reactions

In conventional batch reactor systems, the relatively open reaction space allows for a more unconstrained release of generated gas. The gas evolution rate can be regulated by controlling the feeding rate of materials, thus simplifying the overall operation process.

The confined spatial dimensions of microchannel reactors restrict gas release exclusively through the material outlet. This not only leads to gas accumulation, and a subsequent pressure rise in the reaction system, but also disrupts the fluid flow pattern inside the reactor. The resulting gas slugging effect drastically shortens the residence time of reactants, thereby hindering the completion of the reaction. These issues render gas-generating reactions in microchannel reactors highly challenging. ,, For gas generation, the key solution is to regulate the instantaneous gas pressure generated during the process.

2.1.3.1. Back-Pressure Regulator in Gas-Generating Reactions

BPR serves different functions in systems where gas is generated rather than continuously supplied. It can help maintain pressure stability, prevent gas accumulation, and enhance reaction rates and efficiency. Besides that, BPR ensures that gas will not overflow, and stabilizes flow rates to prevent system instability. Also, BPR improves safety by controlling gas release and preventing hazardous pressure buildup.

2.1.3.2. Spinning Disk Reactor in Gas-Generating Reactions

The gases and reaction liquid in a rotating disc reactor can be discharged from different outlets under the action of centrifugal force. − This unique design allows for efficient phase separation during gas generation processes, where the centrifugal force effectively segregates the gas and liquid phases. As the reactor operates, the centrifugal force causes the reaction liquid to form a thin film or droplets, enhancing the interaction between the gas and liquid phases. The generated gas is then directed toward the gas outlet, while the liquid phase is pushed outward to a separate outlet. This separation ensures that the gas and liquid are effectively removed from the system, preventing the generated gas from entraining the liquid and flushing it out. This mechanism facilitates continuous operation by maintaining a steady flow of both phases, which is crucial for reactions that involve gas generation.

2.1.4. Perspective on Gas–Liquid Reactions

Building on these advancements in reactor design, we believe that incorporating gas recycling and real-time monitoring into flow systems can also be essential for optimizing both efficiency and sustainability. For example, reactor designs can be equipped with real-time gas concentration monitoring systems − that enable precise control over the flow and number of hazardous gases used in reactions. This approach can ensure both safety and stoichiometric accuracy throughout the process. Also, gas recycling systems, like membrane-based separation technologies, allow for the capture and reuse of hazardous or other gases. This provides a way to reduce waste and costs, minimize the environmental impact and enhance process sustainability.

Challenges persist in addressing gas–liquid reactions in continuous flow systems. The rapid and uniform generation and release of gas disrupt the laminar flow-dominated fluid characteristics within the microchannel reactor, leading to the formation of an irregular gas–liquid mixing system and thereby adversely affecting the mixing efficiency of the reaction system. For instance, the continuous release of nitrogen in the widely applied Sandmeyer reaction, and the evolution of hydrogen in the sodium borohydride reduction reaction both encounter the aforementioned challenges. To overcome these, it is noteworthy that integrating the advantages of specialized reactor designs will be crucial. Optimizing flow regimes and reaction conditions in multistep synthesis can further enhance scalability and safety. Future advances could consider focusing on improving reactor scalability, enhancing gas–liquid interaction and process automation to broaden the practical applications of gas–liquid reactions in industrial settings.

2.2. Solid–Liquid Reactions

As early as 2005, an internal evaluation of 86 representative reactions conducted at Lonza revealed that approximately half of these reactions could realize significant economic and operational advantages through a transition from batch to continuous manufacturing, as demonstrated by a comprehensive cost-benefit analysis. Based on kinetic characteristics, the reactions were classified into three distinct categories. Among these three categories, approximately 63% of the reactions involved the presence or utilization of solid-phase components, which means these reactions were not particularly compatible with microreactors at that time. In organic synthesis, many commonly used materials, including ligands, metal catalysts, − and high-molecular-weight organic compounds, are typically obtained or stored as crystalline solids. While small-molecule substrates can often be dissolved and handled as liquids, these crystalline or poorly soluble compounds require specialized strategies for incorporation into flow systems. The presence of insoluble solid substances in a chemical reaction introduces significant challenges for continuous flow operation, as the accumulation of dispersed particulates within the liquid phase can ultimately lead to reactor blockage, commonly referred to as clogging or plugging.

2.2.1. Challenges in Solid–Liquid Reactions

The core issue of solid–liquid reactions lies in the inherently low solubility of the solid phase, which fundamentally limits mass transfer efficiency by disturbing both diffusion and convection within the reaction medium. Solid particles (reactants, intermediates, or byproducts) would create heterogeneous microenvironments. Solute transport to and from the solid surface would be restricted by dissolution kinetics and boundary layer diffusion. And the high surface-to-volume ratio in reactors further accelerates heterogeneous nucleation, leading to bridging, fouling, and settling, which constrict the flow path and alter residence time distributions. As solids accumulate, the effective viscosity of the suspension increases, decreasing fluid velocity and weakening convective mass transfer. Such hydrodynamic instabilities give rise to spatial concentration gradients that compromise uniform reactant distribution and conversion. Ultimately, the interplay between solid phase growth, deposition, and transport resistance transforms what should be a kinetically controlled process into one dominated by diffusion and flow limitations, severely impeding overall reaction performance in continuous systems.

Regarding the issues arising from the presence of solid, they can also be divided into two scenarios to discuss potential solutions, including solid-fed reactions and solid-generated reactions. Strategies focused on these situations in continuous flow systems are discussed in the subsequent sections and illustrated with a set of literature examples.

2.2.1.1. Selecting Solvents or Temperature for Dissolving Solids

Whether in solid-fed reactions or solid-generated reactions, a straightforward strategy to adopt involves reducing concentration, identifying a suitable solvent or changing temperature to dissolve the solid phase. And the solid–liquid reaction can be transformed into a liquid–liquid reaction, thereby minimizing issues related to solid handling and improving process stability. As Baumann, Smyth and co-workers noted, flowability refers to “the assessment of solubility of all starting materials, intermediates, and products under the processing conditions, along with consideration of the stability of feedstocks over the reaction timeframe.”

Wietelmann et al. investigated the continuous processing of organolithium reagents and found that the choice of solvent plays an important role in preventing solid precipitation and reactor clogging. They observed that hydrocarbon solvents such as hexane, which are commonly used in commercial butyllithium solutions (1.6–2.5 M), can act as antisolvents for polar lithiated intermediates. This behavior promotes the formation of insoluble lithium salts, including Li alkoxides, which easily accumulate and block the narrow channels and pathways of microflow reactor systems. To overcome this, they demonstrated that the introduction of THF as a donor solvent effectively enhances the solubility of these polar intermediates. THF disrupts the aggregation of organolithium species, thereby maintaining the homogeneity of the reaction mixture and preventing precipitation. The addition of THF transforms potentially unstable, solid-forming systems into clear, flow-compatible solutions and enables stable and efficient operation in continuous organolithium processes. It is worth noting that THF can react with organolithium reagents, such as tert-butyllithium, under certain conditions, generating ethylene and acetaldehyde lithium enolate, which renders its application somewhat restricted. ,,−

Thaisrivong, Naber and co-workers (Merck Research Laboratories) investigated a continuous Mannich-type addition for the synthesis of verubecestat (MK-8931) and found that temperature could affect the solubility of the reaction components. When the system was cooled below −30 °C, both the starting sulfonamide and the lithium anion formed by deprotonation precipitated, leading to unstable flow and clogging. By increasing the temperature of the heat exchange coils and micromixers to between −10 and 38 °C, the reaction mixture remained homogeneous, achieving steady conversions of 80–87% without blockages. This indicates that moderate temperatures can improve solubility and process stability.

2.2.1.2. Replacement of Feeding Methods

Screening solvents or adjusting the reaction temperature may lead to poorer outcomes. In such cases, modifying the feeding methods can be considered as an alternative approach. The use of a feed pump is a viable solution to consider. Jensen et al. developed a custom-designed slurry pump to address clogging issues in heterogeneous photoredox reactions. The stainless-steel pump was specifically engineered to deliver solid-containing feed stream, like Na₂CO₃ suspension, into the reactor without particle aggregation or channel blockage. It employs a piston-driven hydraulic system that separates the feed from an inert acetonitrile layer, allowing the controlled pressure to drive fluid flow. To maintain a uniform slurry, a magnetic stir bar ensures consistent suspension, while a vibration motor at the outlet enhances solid mobility. Performance testing with an 8.3 wt % Na₂CO₃ slurry in DMF confirmed stable operation with 98.4–103% mass consistency over 100 min. Commercially available peristaltic pumps can also serve as a viable option.

It is worth noting that both the custom-designed slurry pump and commercial peristaltic pump belong to positive displacement pumps, which are widely favored for handling solid-laden slurries due to their reliable solid transportation capability. Their inherent working principle of intermittent fluid displacement may lead to pressure pulsations, a critical but easily overlooked issue that exerts a significant impact on pressure stability and residence time distribution. , To mitigate the pulsations, a common approach to reducing such pulsation is the integration of pressure pulsation dampers downstream of the pump. Bach et al. investigated a bioinspired pulsation damper to mitigate pump pulsations and demonstrated that integrating pressure pulsation dampers downstream of the pump mimics the mechanisms found in biological systems, such as the human circulatory system. Drawing on these biomimetic principles, they developed a bioinspired damper fabricated from cellular rubbers with nonlinear viscoelastic properties. This novel damping solution was tested across a range of operating conditions, including varying back pressures and engine speeds. The results showed that pressure pulsations were reduced by up to 40%.

Standard membrane dampeners may indeed trap solid particles over time, leading to clogging issues, especially when dealing with slurries. Employing membranes with larger pore sizes or coatings that reduce particle adhesion could enhance slurry compatibility. Another potential solution could involve using materials with self-cleaning properties or incorporating dynamic filtering systems that can periodically remove accumulated solids.

Using flow stabilizers like thermally controlled ones and dual pumps driven in different phases are also feasible methods to mitigate pressure pulsations.

2.2.1.3. Inhibition of Solid Accumulation

In addition to the above-mentioned strategies, a few methods can effectively inhibit the settling, fouling, and bridging of solids, thus preventing clogging.

(i) Solid-Tolerant Microreactor. When designing or selecting a microreactor capable of handling solids, the primary strategy is to ensure uninterrupted flow by minimizing solid particle accumulation. A direct and effective approach is to increase the internal diameter (ID) of the reactor channels, as wider tubing reduces solid aggregation and lowers the risk of bridging between solid particles. ,

Ley et al. addressed the problem of clogging in the continuous synthesis of SR48692 by modifying the internal diameter of the flow reactor tubing. In the initial setup, blockages frequently occurred due to precipitation of the Claisen condensation product when using narrow tubing. And the reaction failed to reach completion when the concentration was lower. To resolve this, they expanded the perfluoro alkoxy alkane (PFA) coil to a wider one with a 52 mL volume, which effectively avoided the obstruction.

(ii) Mixers. The development of mixers in continuous flow chemistry has evolved from simple mixer designs to more complicated ones. The relatively simple mixers are T- and Y-piece mixers. − In addition, V-shaped mixer has also been used. But T-, Y- or V- mixers can only provide simple mixing and are not tolerant of reactions that generate solid phases.

In addition to these mixers, static mixers, which feature a specialized design with fixed, stationary internal components, can significantly enhance mixing efficiency. They can serve as a feasible option for slurry-forming reactions that demand strong mixing performance. Static designs remain limited in slower reactions, handling suspensions or crystallizing reactions, as solids tended to accumulate on internal surfaces. ,

To address these issues, Dolman et al. developed a flow mixer called the magnetically driven agitation in a tube (MDAT) mixer. It was designed to achieve rapid and clog-resistant mixing independent of flow rate. Structurally, the MDAT consists of a short section of stainless-steel tubing containing two small magnetic stir bars, positioned over a standard magnetic stir plate. The tumbling motion of the magnets inside the tube generates strong local turbulence, thereby enhancing mixing efficiency. In their experiments, Dolman et al. demonstrated that the MDAT provided superior mixing performance compared to both T- and multilaminar mixers, especially under low-flow or cryogenic conditions. It effectively prevented clogging even during organolithium and electrophilic quench reactions at −78 °C, allowing complete conversion without requiring high dilution.

Inspired by MDAT mixer, a dynamic mixer was designed. Dynamic mixer typically features a rotor-stator configuration or rotating elements within a tube, such as cone-shaped rotors, hemispherical cavity systems, or magnetically driven stir bars. These components generate active motion, producing localized turbulence and shear zones that continuously disrupt laminar flow. Compared with the static mixer, a dynamic mixer can tolerate a higher solid content because of its active agitation that prevents particle settling and accumulation.

Calvin et al. introduced a magnetically driven dynamic mixer between the zincate generation and Negishi coupling modules to overcome clogging and mixing issues caused by solid formation during continuous operation. In this setup, a polytetrafluoroethylene (PTFE) tube containing a small magnetic stir bar was placed on a magnetic stirrer, generating continuous agitation as the feeds entered. This motion ensured thorough blending of the zincate reagent and aryl bromide-palladium catalyst solution. Solid particles remained suspended rather than accumulating, effectively avoiding blockages and maintaining uniform flow. The Negishi coupling combined the zincate with quinoline bromide and 1 mol % palladium tetrakis (triphenyl phosphine) at 60 °C for 1 h, achieving full conversion.

(iii) Packed-Bed Reactor. The packed-bed reactor operates by passing reactants through a stationary bed filled with solid reactants. It allows reactions to occur on the solid surface as fluids flow either downward (trickle-bed mode) or upward (flooded-bed mode), which maximizes the contact area between the gas, liquid, and solid phases. Usually, it consists of packed catalyst particles with controlled porosity and particle size, like palladium catalysts, ,− nickel catalysts, , and biocatalysts. Alternatively, it can also be packed with solids that participate in the reaction.

The Ley group synthesized (±)-oxomaritidine 2–23 using a continuous flow system composed of a series of packed columns containing immobilized reagents, catalysts, and scavengers (Figure A). The sequence began with 2–15, which was converted to an azide 2–17 by passing through an azide exchange resin column. In parallel, 2–16 was oxidized to an aldehyde 2–18 using a prepacked column of tetra-N-alkylammonium perruthenate. Subsequently, an aza-Wittig reaction was performed on a polymer-supported phosphine-packed column, to form an imine 2–19, and then it was hydrogenated in a 10% Pd/C cartridge to yield a secondary amine 2–20. After solvent exchange, the amine was protected with a trifluoroacetyl group in a microreactor and then passed through a silica-supported primary amine, removing any excess trifluoroacetic anhydride (TFAA) or residual trifluoroacetic acid (TFA). The reaction stream was subsequently directed through a polymer-supported PIFA column, which mediated an oxidative phenolic coupling, forming a seven-membered tricyclic intermediate 2–22. This product mixture was then passed directly into a polymer-supported basic resin column, where amide bond cleavage and intramolecular 1,4-conjugate addition occurred spontaneously to yield (±)-oxomaritidine 2–23. The entire multistep process was conducted continuously with each packed-bed reactor performing a distinct transformation step in sequence. The overall isolated yield reached approximately 40%, which is equivalent to batch processes.

4.

(A) Continuous flow synthesis of (±)-oxomaritidine (2–23) by the Ley group. THF: tetrahydrofuran. PIFA: [bis­(trifluoroacetoxy)­iodo] benzene. TFAA: trifluoroacetic anhydride. (B) A cascade of CSTRs for preparing edivoxetine HCl intermediate (2–27) by Kopach and co-workers. (C) Continuous photocatalytic reaction catalyzed by TiO₂ particles run in meso-scale ultrasonic milli-reactor by Noël et al.

(iv) Continuous Stirred Tank Reactor (CSTR). The consists of a fed vessel, which is equipped with an individual agitator. And it can accommodate solids by employing continuous stirring and efficient mixing. − This design can keep solid particles uniformly suspended and prevent their sedimentation or accumulation within the reactor. In practical applications, multiple cascaded CSTRs tend to be used more frequently than a single CSTR. The use of multistage cascades can overcome the problems of a single CSTR, namely broad residence time distribution and the resulting deterioration in volumetric throughput and reaction timing control. CSTRs are now extensively used in various applications, like Grignard reactions. − Kopach and co-workers developed a continuous Barbier reaction using a cascade of multistage CSTRs to synthesize a key intermediate for edivoxetine·HCl production (Figure B). The process employed a series of CSTRs. Continuous synthesis of a Grignard reagent 2–25 from 2–24 was achieved in the first CSTR. Subsequently, an addition reaction between 2–26 and 2–25 takes place in the second CSTR. After acidic quenching in the third CSTR and liquid–liquid extraction, the desired intermediate 2–27 was obtained. CSTRs can also be applied to many other solid-involving reactions ,, and still hold considerable potential for further development.

(v)­Ultrasound in Solid–Liquid Reactions. The acoustic cavitation generated by the ultrasound can promote localized turbulence, enhance mass transfer, and help to prevent solid deposition or channel blockage during reactions. The application of ultrasound in flow chemistry originated from the use of ultrasonic baths, which represented the earliest attempt to introduce acoustic energy into continuous systems.

Also, this represents a relatively simple and straightforward way to apply ultrasound. In these setups, the entire flow reactor composed of coiled tubing or microchannels, was immersed in an ultrasound bath. −

Noël et al. designed a meso-scale ultrasonic milli-reactor to perform a gas–liquid–solid photocatalytic oxidation of alcohol 2–28 using TiO₂ particles as the photocatalyst (Figure C). The reactor consisted of a Langevin-type ultrasonic transducer connected to a titanium sonotrode, which was attached to a glass reactor coil irradiated by a UV-A light (365 nm) source. The ultrasonic field, operated at 22.6 kHz and 60 W, generated cavitation and acoustic microstreaming within the flow, producing vigorous turbulence and dispersing the TiO₂ particles evenly throughout the reaction medium. The ultrasound was applied in a pulsed mode (ON/OFF time of 7.5/12.5 s) to avoid excessive heating while maintaining strong mixing and uniform suspension. During operation, the system effectively prevented particle settling and ensured stable continuous flow of the three-phase reaction mixture.

Another approach to applying ultrasound is the use of reactors equipped with built-in ultrasonic transducers. Ultrasonic reactors can integrate piezoelectric or Langevin-type transducers directly into the reactor body, thereby producing a more uniform and focused cavitation field. Ultrasonic reactors can allow localized and programmable sonication, reducing unwanted heating and improving reproducibility. ,,

The compact and modular design of ultrasonic reactors also makes them easier to integrate into microreactor systems and multistep continuous flow systems.

(vi) Spinning Disk Reactor (SDR) in Solid–Liquid Reactions. The spinning disc reactor in solid–liquid systems operates based on the principle of intense centrifugal force and thinning/diversing-film flow generated by a rapidly rotating disc. − The main difference lies in the modification of the gas inlet into a solid feed port, allowing suspended solids or precipitated particles to be evenly distributed within the rotating liquid film without deposition. Under high rotational speeds, the solid particles are subjected to intense shear forces within the thin liquid film. And this can keep them uniformly suspended, prevent aggregation or adhesion to the disc surface, and facilitate their rapid formation and dispersion throughout the reaction medium. This design has proven highly effective in reactions including but not limited to organometallic reactions, photochemical reactions, and diazotization reactions, with excellent solid tolerance.

(vii)­Multiphase Flow. Multiphase flow patterns can be employed to further enhance mass and heat transfer efficiency in multiphase systems.

In slug or droplet flow systems, alternating segments of immiscible fluids or inert gases create internal recirculation vortices within each slug or droplet, significantly improving mixing and surface renewal at phase boundaries. −

Zhang, Tang, Wu and co-workers developed a continuous gas–liquid–solid slug flow system to improve solid–liquid photoinduced electron/energy transfer reversible addition–fragmentation chain transfer polymerization and prevent clogging. They used nitrogen gas to segment the slurry of methyl methacrylate and the solid photocatalyst into discrete slugs within a PFA capillary reactor. The continuous slurry flow was segmented into a series of microbatch reactors. Within each liquid slug, the inert gas induced internal recirculation vortices, maintaining uniform suspension of the solid under LED irradiation (360–370 nm, 500 W, 29.1 mW/cm²). The inert gases also acted as a barrier against particle sedimentation and channel fouling.

Alternatively, a monomer solution can be utilized as a substitute for inert gases in certain cases. −

2.2.1.4. Continuous Crystallization

Continuous crystallization, as a special type of solid–liquid reaction, is widely applied in the industry, ranging from the production of common bulk chemicals to highly specialized and high-value compounds. For example, it holds significant importance in the pharmaceutical field, where over 90% of active pharmaceutical ingredients (APIs) are produced or isolated in crystalline form. ,

Continuous crystallization can be optimized by adjusting key process parameters, including residence time, the number of stages or crystallizers, and the temperature as well as antisolvent addition rate in each stage. In addition to adjusting key process parameters, process intensification through reactor design optimization remains one of the most fundamentally direct and widely adopted strategies for enhancing the performance of continuous crystallization.

The multistage mixed suspension mixed product removal (MSMPR) crystallizers cascade is one of the most widely adopted designs in pharmaceutical crystallization. Like CSTRs, by connecting several well-mixed tanks in series, each stage can be tailored to a specific process function, such as nucleation, crystal growth, or aging. This can allow independent control of supersaturation and residence time at each step and minimize the broad residence time distribution typically observed in a single crystallizer, which means it can produce narrower crystal size distributions and more consistent product quality. Myerson et al. implemented a two-stage MSMPR cascade crystallization system for the manufacture of ciprofloxacin HCl. The process began with pH-neutralization precipitation of crude ciprofloxacin and formed a suspension in the first MSMPR reactor. The slurry was continuously transferred to the second MSMPR unit for steady-state operation. After purification, the API solution was subjected to antisolvent crystallization in another two-stage MSMPR setup, where isopropanol acted as the antisolvent. Crystallization proceeded at 5 °C with 56 min residence time per stage and produced ciprofloxacin HCl hydrate with 98.80 ± 0.20% purity and 58.8 ± 7.7% yield.

Several other types of crystallizers are also available, like tubular or plug flow crystallizers (PFC), laminar-flow tubular crystallizers (LFTC), coiled flow inverter (CFI) crystallizers, segmented/slug flow crystallizers, and continuous oscillatory baffled crystallizers (COBC).

Other emerging strategies have been developed to further enhance crystallization performance and product quality, which can also serve as viable considerations for us. Techniques like process analytical technology (PAT) integration and model-based predictive control (MPC) can offer new avenues for process intensification and stability improvement. −

2.2.1.5. Perspectives on Solid–Liquid Reactions

Grinding solid materials down to the nanoscale by mechanochemistry offers a promising solution for improving the efficiency of solid–liquid reactions. By reducing particle size to the nanometer range, nanoscale particles can exhibit shorter diffusion paths and higher defect densities. It can accelerate reaction kinetics and lower activation barriers. When combined with continuous flow systems, such pretreated nanostructured solids can remain more uniformly suspended, minimizing sedimentation and clogging. ,

Fundamentally, most existing strategies for advancing solid–liquid reactions are intended to optimize solid dispersion and interfacial mass transfer. The appropriate solution is not universal across different reactions, as the challenges vary depending on factors.

Looking ahead, the future development of solid–liquid flow chemistry will likely focus on integrating multiscale control and adaptive reactor design. Also, the incorporation of optimization, driven by artificial intelligence and modeling assisted by data, will allow predictive control of solid behavior and phase transitions. This, in turn, enables robust, scalable, and sustainable operation of solid–liquid processes across diverse chemical manufacturing situations.

We believe that in the future, the handling of solid–liquid reactions will become more accessible and user-friendly. Gas–liquid–solid three-phase reactions possess the core characteristics of both gas–liquid and solid–liquid systems, yet their complexity is significantly higher than the latter two. While the solutions to their challenges are essentially the integration of strategies applicable to the aforementioned two-phase systems, relevant details can be found in the preceding discussion on the two-phase systems, and no further elaboration is provided herein.

2.3. Reaction Compatibility Matching

2.3.1. Challenges in Reaction Compatibility Matching

The main issues of reaction compatibility matching lie in the optimization of individual reaction steps and the integration between individual modules.

Individual reaction steps serve as the cornerstones for multistep continuous synthesis. It is essential to ensure that each individual reaction achieves sufficiently high purity, conversion, and yield. Also, proper coordination of reaction conditions including solvents, reagents, and their usage, as well as downstream processing, should be ensured by the transitions between different reaction steps.

The specific situations and possible solutions related to reaction compatibility matching will be discussed in the following section.

2.3.1.1. Compatibility among Various Systems

(i) Compatibility of Reaction System. For the reaction system, the first consideration should be the compatibility of all substances involved in the reaction. And even the materials of the tubes that come into contact with the solution should also be considered. For example, the dimensional parameters and material composition of the tubes are significantly important due to their direct interaction with the reagent flow stream. Key physical factors, such as inner diameter and wall thickness of the tubing and chemical compatibility between the tubing and reagents demand careful consideration during the design phase. Generally, inert perfluorinated polymers, such as PTFE, PFA, polyetheretherketone (PEEK), fluorinated ethylene propylene (FEP), are adequate. Chemical compatibility refers to the requirement that all components present in an upstream reaction system do not affect subsequent steps. Common incompatibility-related problems include corrosion of tubing materials or metal distributors induced by highly reactive intermediates, pH fluctuation or trace impurities that give rise to downstream catalyst deactivation, and entrainment of bases or oxidants which can decompose reagents or intermediates in subsequent reaction steps. When chemical compatibility cannot be ensured, appropriate separation or purification steps are required before proceeding to the next stage.

In the continuous manufacturing of rocuronium bromide (Figure A), Neshchadin, Hu, Mascia and co-workers found that in the acetylation step, the compatibility between the AcCl and the base had a direct impact on impurity formation (Figure B).

5.

(A) Synthetic route of rocuronium bromide by Neshchadin, Hu, Mascia and co-workers. (B) Impurity formation of acetylation step by Neshchadin, Hu, Mascia and co-workers. (C) Continuous flow synthesis involving solvent switch using microfluidic distillation by Buchwald, Jensen and co-workers. DIPEA: N,N-diisopropylethylamine. AcCl: acetyl chloride. TEA: triethylamine. Pd­(OAc)₂: palladium­(II) acetate.

The generation of the 3-acetylated impurity (2–33) and the diacetylated byproduct (2–34) was highly sensitive to the matching degree of these two reagents. It was essential to select a base that could suppress side reactions while promoting the selective acetylation of the hydroxyl group on MPAD (2–30). Among the screened candidates, DIPEA demonstrated superior compatibility with AcCl compared to TEA. When 1.3 equiv of DIPEA were used, complete conversion of MPAD was achieved, yielding AMPO (2–31) with 95.23% purity and only 1.11% diacetyl impurity. Under identical conditions, TEA resulted in 0.89% residual MPAD and an increased 3.37% diacetyl impurity, confirming that DIPEA was the more suitable and compatible base for continuous acetylation.

(ii) Compatibility of Solvent System. For the solvent system, the ideal scenario is to use a single, unified solvent throughout all reaction steps. Achieving this is often challenging, as it places significant constraints on both the reaction conditions and the optimization of each individual step. ,

The use of mixed solvent systems or in-line solvent swaps has become the primary strategy to address these compatibility challenges. ,,− Buchwald, Jensen and co-workers resolved solvent incompatibility between sequential reactions by implementing microfluidic distillation for in-line solvent exchange (Figure C). The aryl triflate (2–36) was synthesized in DCM, which was incompatible with the subsequent palladium-catalyzed Heck coupling that required a higher-boiling solvent such as toluene or DMF. To address this, the product stream from the first reaction was directed into a microdistillation unit. In this unit, gas–liquid segmented flow and temperature control could enable selective evaporation of DCM while retaining the product and replacing it with the new solvent. After the in-line solvent exchange, the stream was directly introduced into the Heck coupling module and yielded the desired product 2–38 in high purity and achieved a total yield of up to 76.8%.

2.3.1.2. Temporal Compatibility

In multistep continuous flow reactions, particularly end-to-end flow processes, ensuring smooth operation of the reaction system requires well-coordinated residence times across every microreactor and each workup unit. Nevertheless, the rates of different chemical reactions always vary significantly in practice. For example, some reactions reach completion within seconds, whereas others may require several minutes or even hours to achieve full conversion. This presents challenges for ensuring temporal compatibility in multistep continuous flow reactions. Strategies are often required to ensure the time consistency between multistep reactions and commonly adopted approaches are discussed below.

A straightforward strategy is to optimize the reaction conditions, such as increasing the reaction concentration or raising the temperature, thereby accelerating the reaction. Neshchadin, Hu, Mascia and co-workers found that in the N-alkylation step, the reaction time could be reduced from 24 h to just 6 h by decreasing the solvent volume from 25 vol to 5 vol relative to the substrate (Figure A).

Reactors, as the core units of every flow system where chemical reactions take place, can be replaced with types featuring higher mixing efficiency. The reaction mixture can be circulated within the reactor for extended reaction times, while faster-reacting solutions can be temporarily stored in holding tanks, allowing coordination and overall temporal compatibility across all reaction steps. ,−

2.3.1.3. In-Line Workup

Numerous reactions still necessitate workup to quench the reaction or remove side products. To ensure the continuity of the process, in-line workup is indispensable in most multistep continuous flow syntheses.

(i) Liquid–Liquid Separation. Currently, a frequently employed method in this field is membrane-based liquid–liquid separation (i.e., liquid–liquid extraction). Its operating mechanism is that an extractant is added to the reaction stream via a mixer to form a biphasic mixture, which then flows through a holding unit for extraction. Subsequently, the mixture enters a membrane separator with a PTFE membrane between two channels. The organic phase permeates the hydrophobic membrane under pressure control, and both phases are available for subsequent steps. ,−

The gravity-based extractor, which resembles traditional mixer-settlers, can also be used as another extraction method. ,

(ii) Gas–Liquid Separation. As for gas–liquid separation, if the outlet is directly open to the atmosphere, the gas can readily escape from the solution. In such cases, the use of a gas–liquid separator is unnecessary.

If the outlet is not directly open to the atmosphere, or if the system is sealed or closed, gases cannot escape freely from the solution and may accumulate within the reactor or downstream flow line. Under such circumstances, similar to liquid–liquid separation, membrane-based separators − and gravity-based separators , can also be used for gas–liquid separation. ,

(iii) Distillation. Distillation, as a specialized form of gas–liquid separation, is primarily applied for in-line solvent exchange in continuous processes. The common implementation methods can generally be categorized into the following several strategies, namely vacuum distillation, capillary force distillation, centrifugal force/gravity effect distillation and other types of distillation. It is also a viable strategy to prolong the residence time of the reaction solution, which can be achieved by increasing the number of reactors, expanding the holding volume, or reducing the flow rate.

In vacuum distillation, the system pressure is reduced to lower the boiling points of liquid components, allowing separation to occur at mild temperatures. Within microscale setups, channels with a specially designed shape, such as serpentine ones, , and integrated heating–cooling zones promote rapid vaporization and condensation. This strategy is suited for heat-sensitive compounds. −

Capillary distillation relies on surface tension and capillary forces to drive liquid motion through micro- or nanostructured channels, instead of mechanical pumping. A stable vapor–liquid interface is formed by the process, where selective evaporation and condensation occur under continuous flow. ,−

Sharing similarities with SDR, centrifugal force/gravity effect distillation enhances phase segregation through rotational or gravitational acceleration. In rotating spiral channels, the centrifugal forces the vapor and liquid phases to migrate in opposite radial directions according to their density differences. The continuous separation of the lighter vapor from the heavier liquid establishes a stable gas–liquid interface, facilitating rapid phase segregation, while the intensified centrifugal force effectively accelerates droplet aggregation and vapor release. ,,−

Other approaches have also been developed, such as bubble-assisted interphase mass transfer, which enhances evaporation through localized heating near gas–liquid interfaces, and polymer multilayer microchips, which utilize microchannel arrays to control vapor diffusion and condensation. , Ley et al. designed a prototype evaporation device capable of operating in both batch and continuous modes. The system combines precise temperature and pressure control within a sealed chamber. Real-time solvent evaporation or exchange during continuous processing is enabled without decomposition or product loss, while operational flexibility and process efficiency are greatly improved.

(iv) Solid–Liquid Separation. Solid–liquid separation is an essential technique in in-line processing, enabling filtration and solid recovery, thereby facilitating continuous product purification, catalyst recycling, and overall process intensification in continuous flow systems. Current filtration methods can be primarily categorized into filter-based and filterless systems. −

As for the filter-based system, an extensively employed approach is membrane-based filtration, such as nanofiltration. − It operates by using semipermeable membranes to separate components based on size or molecular weight, allowing smaller molecules to pass through while retaining larger particles or solutes. In the first step of synthesizing tiadinil and methiadinil, which was conducted in batch mode, MgSO₄ used for drying would cause clogging of the microreactor. Based on the membrane filtration method, we performed filtration through a membrane-equipped chromatographic column connected with a plunger pump to remove MgSO₄ before feeding the material into the subsequent reaction chip.

The packed-bed filter is also a viable option, which uses a bed of granular material to capture suspended solids when the liquid passes through. Structurally, the granular material is packed inside to form a bed that captures suspended solids during liquid passage. −

There are also various filters made from various materials available for continuous filtration, like stainless steel, which utilize the porous structure of the material to achieve solid–liquid separation. And the liquid passes through the pores while solid particles are trapped, thereby separating the solids from the flow.

As for the filterless system, it primarily relies on forces or auxiliary energies to achieve solid–liquid separation. For example, the knotted reactor uses secondary centrifugal forces generated by the reactor’s three-dimensional coiled structure to adsorb precipitates onto the pipe wall, achieving solid–liquid separation. − And ultrasound-assisted filtration uses ultrasonic standing waves to aggregate particles at the center of the flow channel, with clarified liquid flowing out from the wall. − Additionally, there are electro/magnetic-assisted filtration methods, mainly used in the biological field, such as electrofiltration for separating cationic peptides and magnetic filtration for separating protein-containing particles.

The multistep continuous flow for the preparation of artemisinin and derivatives (Figure A) by Seeberger et al. integrated a fully continuous sequence of gas–liquid, liquid–liquid, and solid–liquid separations to ensure uninterrupted operation and high product purity (Figure B). Initially, 2–40 underwent photooxidation to artemisinin in a gas–liquid photoreactor, where oxygen was introduced and subsequently separated from the liquid phase using a membrane-based gas–liquid separator. The resulting organic stream underwent reduction in the packed-bed, followed by separation of the gas phase through a gas–liquid separator and the water phase through a membrane-based extractor, and yielded 2–42. Then derivatization to 2–43, 2–44, or 2–45 was performed in a biphasic system, where another membrane-based extractor continuously isolated aqueous and organic phases. Also, 2–45 could be purified as a three-stage solid–liquid process, including precipitation and filtration after n-hexane dilution, multicolumn chromatography for impurity removal, and continuous crystallization for final product isolation. And compounds 2–45 were finally obtained with a purity exceeding 99.5%.

6.

(A) Most common artemisinin precursors and derivatives. (B) Continuous flow preparation of artemisinin and derivatives by Seeberger et al. DCA: 9,10-dicyanoanthracene. EtOH: ethanol. TMOF: trimethyl orthoformate. TEOF: triethyl orthoformate. SAA: succinic anhydride.

2.3.2. Perspective on Reaction Compatibility Matching

We believe that, with the increasing availability of techniques to adjust reaction compatibility, future research should place more emphasis on the chemical reactions themselves rather than on compatibility concerns. And the long-term stability and continuity of multistep continuous flow processes will become a crucial focus for future development. In addition, newly emerging reactions and methodologies, such as C–H bond activation − and catalytic asymmetric dearomatization (CADA) reactions, − should gradually be integrated into multistep flow systems, with efforts directed toward process development and optimization to meet synthetic demands.

2.4. Scale-Up Issues

2.4.1. Challenges in Scale-Up

Flow chemistry enables seamless scale-up from laboratory to industrial production. It benefits from superior heat and mass transfer efficiency, an enhanced safety profile, and precise control over reaction parameters, which are often difficult to achieve in conventional batch or semibatch reactors. − A challenging problem is that the unique microscale transport and control that make flow chemistry so powerful at small scale may not be faithfully preserved when the reactor is scaled to industrial throughputs. There are two main strategies for scaling up flow reactions, including numbering up and sizing up.

2.4.1.1. Numbering up

Numbering up is widely used because it allows higher throughput by operating multiple identical microreactors in parallel while preserving the favorable characteristics of microscale operation. This approach maintains excellent heat and mass transfer, uniform reaction environments, and high process safety, as each reactor operates under the same hydrodynamic and transport conditions as the laboratory-scale system. Thermal effects are relatively easy to control in numbering up systems, given that each unit maintains a modest individual scale, making thermal runaway less likely to occur. Heat transfer is tightly linked to thermal runaway risk. For exothermic reactions, effective heat removal is critical, yet convective heat transport with flowing fluid outperforms static conduction, which creates potential hazards if flow stops. The major challenge in numbering up lies in achieving uniform flow distribution among all channels to ensure consistent residence times and reaction performance. Various flow distributor designs, such as chamber, bifurcation, split-and-recombine, have been developed to minimize maldistribution and pressure drop. − The numbering-up approach is also challenged by its high cost. Fabricating numerous identical microreactors, assembling complex manifolds, and ensuring precise flow distribution will increase both capital investment and operational expenses, making scale-up implementation economically challenging.

2.4.1.2. Sizing up

Sizing-up strategy focuses on enlarging the physical dimensions of individual reactors, like channel diameter or length, to increase production capacity within a single unit. By operating one larger reactor rather than multiple small ones, this approach simplifies system integration and reduces the complexity of flow distribution networks.

Scaling up the channel dimensions inevitably alters the hydrodynamic and transport properties that define microreactor performance.

As the surface-to-volume ratio decreases, both heat and mass transfer efficiencies may decline. As a result, nonuniform concentration profiles and reduced reaction selectivity may be induced. To mitigate these drawbacks, strategies have been developed, such as geometric similarity, constant pressure-drop design, and the incorporation of static or dynamic mixing structures. ,,−

Unlike numbering up, sizing up is often accompanied by significant pressure drop and thermal runaway and mixing will become less efficient. Increasing the channel diameter while only moderately extending the channel length can be implemented to maintain a constant pressure drop, which effectively reduces flow resistance at a given throughput. And efficient internal mixing structures such as static mixers, helical or curved channels, and flow inverters are introduced to generate secondary flows and thereby enhance mixing and heat transfer. Enhanced heat removal through high-conductivity reactor materials and intensified heat exchange can suppress local hot spots and reduce the likelihood of thermal runaway. Real-time monitoring and feedback control of pressure, temperature, and flow rate enable early detection of deviations from stable operation, allowing timely adjustments and ensuring safety.

Maintaining the same level of process intensification and safety as in microreactors remains a major challenge. McMullen and co-workers scaled up an aldol condensation reaction used in the synthesis of an API intermediate by employing a sizing-up strategy (Figure ). The process was first optimized at the milliliter laboratory scale using a flow reactor composed of precooled reagent streams (Figure A). After addressing issues such as line plugging by optimizing solvent composition with 10% THF, they focused on matching flow regimes rather than identical Reynolds numbers to ensure consistent mixing performance across scales.

7.

(A) Laboratory flow process for continuous aldol formation with subsequent diene formation by McMullen and co-workers. (B) Pilot plant process flow diagram for continuous aldol formation with subsequent diene formation by McMullen and co-workers. KOtAM: potassium tert-amylate.

For pilot-scale production, the reactor was enlarged to a Y-mixer made of 1/4-in. Hastelloy-C SCH- 40 pipe (9.2 mL for mixed stream volume) connected to modular heat exchangers (Figure B). The total flow rate was increased from 11.2 mL/min in the lab to 1.6 L/min in the pilot plant and achieved a transitional flow regime similar to laboratory conditions. The scaled-up continuous reactor processed about 200 kg 2–47, maintaining stable temperature and pressure profiles with no evidence of fouling or plugging. Finally, the isolated yield of 2–50 was 68%, consistent with laboratory results.

2.4.2. Perspective on Scale-Up Issues

We believe that translating flow chemistry from the laboratory to industrial production will become increasingly straightforward with continued technological advancements. In our opinion, developing predictive models to simulate and optimize microreactor scale-up behavior will play a crucial role in bridging this gap. In the future, the integration of machine learning and real-time process analytics is expected to revolutionize scale-up strategies, enabling automated design and cost-efficient production across scales.

2.5. AI-Assisted Flow Synthesis

2.5.1. Challenges in AI-Assisted Flow Synthesis

Currently, artificial intelligence (AI) has made remarkable progress in organic synthesis, especially in retrosynthetic route design, where its involvement is relatively mature. Through deep learning and machine learning algorithms, AI can identify reaction patterns from extensive literature and databases to predict conditions and propose viable synthetic routes. − Nevertheless, when extending these capabilities to flow chemistry, considerable challenges persist for lack of high-quality data sets that are indispensable for model training.

Reaction data sets often lack detailed specific parameters for flow chemistry, such as reactor geometry, residence time distribution, and mixing efficiency, which significantly limit the accuracy of predictive models. The diversity of hardware types and the absence of real-time feedback mechanisms further hinder the optimization of reactions in closed-loop systems, posing substantial challenges to AI’s ability to learn and adapt dynamically in flow chemistry systems. − Although high-throughput platforms could be employed to generate vast amounts of data, the data generated by different laboratories lacks unified specifications, and such data fragmentation makes it difficult to be utilized for AI learning and training. To overcome these barriers, it is essential to establish a standardized metadata schema specifically designed for flow chemistry.

For instance, standardizing the reporting details of experimental details such as reactor geometry or mixing efficiency would ensure that data from one hardware platform can be easily transferred and utilized on another.

Strategies have been proposed and implemented to address these challenges, including the development of standardized data formats and the establishment of open-access databases like the Open Reaction Database (ORD) developed by Kearnes, Coley and co-workers. The ORD scheme provides a structured framework that accommodates both traditional benchtop reactions and flow chemistry systems, capturing essential experimental details such as inputs, conditions, and outcomes. By ensuring that data is reported in a consistent, machine-readable format, these efforts aim to enhance the accuracy of AI models and enable more effective closed-loop optimization of flow chemistry reactions.

2.5.1.1. Process Analytical Technology

Process analytical technology (PAT) is a framework that integrates analytical tools, sensors, and control strategies to monitor and optimize chemical or pharmaceutical manufacturing processes in real time. By employing analytical tools, like inline IR (infrared spectroscopy), − inline NMR (nuclear magnetic Resonance spectroscopy), − online UPLC (ultra performance liquid chromatography), − online Raman spectroscopy detection − and online UV/vis (ultraviolet/visible spectroscopy), , reaction process data can be continuously and accurately collected in real time. And a reliable basis can be provided for subsequent analysis and process optimization.

Kappe and co-workers developed a modular continuous-flow platform integrating multiple PAT tools including inline IR, inline NMR, and online UPLC to enable real-time monitoring and optimization of multistep reactions (Figure A). As a demonstration, a three-step organolithium transformation was continuously monitored at several reaction points (Figure B). Inline IR spectroscopy observed the disappearance of 2–51’s ester carbonyl signal during enolate formation, while inline NMR tracked the consumption of 2–53 and the formation of intermediates 2–54. And detailed compositional data for the final reaction mixture was provided by online UPLC, which was equipped with automated subsampling and dilution. All analytical instruments were digitally connected to a LabVision control system, allowing synchronized data collection and process feedback. Under optimized conditions, the system achieved stable continuous operation with product yields ranging from 70% to 85% and a productivity of 4.2 g/h.

8.

(A) A three-step organolithium transformation. (B) A modular flow platform with multiple integrated PAT tools for a three-step organolithium transformation by Kappe and co-workers. LDA: Lithium diisopropylamide.

AI is not only applied to reaction prediction and outcome evaluation, but also plays a role in the safety monitoring of reaction processes, particularly in the early warning of sudden clogging risks. Clogging can be monitored from multiple aspects by leveraging PAT signals, such as pressure, optics, image signals, acoustics and viscosity. For the pressure aspect, pressure sensors are used to monitor trends in pressure changes. Optical approaches can employ light transmittance indicators to determine system turbidity. The real-time image data is utilized by AI tools to identify potential clogging risks. For the acoustic aspect, methods such as dynamic ultrasonic scattering are employed. The change of viscosity of the reactant could also be applied to predict potential clogging.

Coupling specific PAT signals with data-driven models, algorithms, and neural networks to identify precursors of clogging evolution has emerged as an increasingly mature approach in continuous flow system monitoring. , We believe that by establishing quantitative links between PAT signals and clogging mechanisms, hybrid frameworks integrating physics-informed constraints and learning-based models can be developed, enabling earlier detection, reduced false alarms, and more robust deployment in real industrial operating conditions.

2.5.1.2. Algorithms for Self-Optimization

Based on high-throughput experimentation − and PAT, algorithms for self-optimization have gradually been developed to enhance continuous flow chemistry systems. Currently, three primary algorithm categories are commonly utilized in self-optimization experimentation, namely local optimization algorithms including design of experiments (DoE) , and the Nelder–Mead simplex, , global optimization techniques such as SNOBFIT , and Bayesian Optimization, ,,− and machine learning algorithms exemplified by deep reinforcement learning. ,

Jensen et al. are pioneers in this emerging field and they have developed diverse automated continuous flow platforms. ,,, A multistep synthetic route for sonidegib has been optimized using a modular robotic flow platform integrated with a multiobjective Bayesian optimization algorithm by Jamison, Jensen and co-workers (Figure ). The process began with computer-aided synthesis planning to generate a retrosynthetic pathway and recommend initial reaction conditions. These proposed conditions were refined manually, considering solubility and reactivity issues that arose during early tests. The first reaction in the synthesis, a nucleophilic aromatic substitution (S_NAr) reaction, was optimized using the Bayesian algorithm. This algorithm handled both continuous variables such as temperature and residence time, and categorical variables such as the halide leaving group (Cl, Br, or F). Following the S_NAr step, the reduction of 2–58 was optimized next. The reduction was carried out in a packed-bed reactor, and the optimization focused on controlling the temperature and flow rates to enhance the conversion and throughput. Subsequently, the amide coupling step was optimized using HATU as the coupling reagent, which was chosen after a comparison with other reagents like EDC/HOBt. The optimization considered both continuous variables like reaction time and temperature, and discrete variables like the reactor volume (1 or 3 mL). The optimization process also allowed the platform to adjust residence times by switching between reactor volumes. After optimizing the multistep synthesis of sonidegib, the final reaction conditions achieved a yield of 93% with a productivity of 7.4 g/h.

9.

Multistep continuous flow synthesis of sonidegib by Jamison, Jensen and co-workers driven by the Bayesian optimization algorithm and implemented on an automated robotic platform. HATU: hexafluorophosphate azabenzotriazole tetramethyl uronium. EDC: 1-ethyl-3-(3-(dimethylamino)­propyl) carbodiimide.

Bayesian optimization has been applied to single-step reaction optimization and kinetic parameter estimation for both macro- and microkinetic models, particularly in black-box settings where gradients are unavailable or simulations are expensive. Existing studies mainly focus on optimizing reaction conditions or fitting parameters for isolated steps, rather than explicitly addressing multistep reaction kinetics. , Multistep kinetic systems introduce additional challenges, such as parameter interdependence and error accumulation across steps. Drawing inspiration from current Bayesian optimization frameworks and tools, such as petBOA, future work could extend Bayesian optimization methodologies to multistep kinetics by incorporating mechanistic constraints, sensitivity analysis, and hierarchical or multifidelity modeling, enabling efficient and physically consistent optimization of complex reaction networks. In addition to Bayesian algorithms, other algorithms offer unique advantages in different research scenarios such as convolutional neural network, recurrent neural networks etc., and are worthy of researchers’ consideration based on specific needs.

2.5.1.3. Digital Twins

The digital twin is defined as “a set of virtual information constructs that mimics the structure, context and behavior of an individual or unique physical asset, that is dynamically updated with data from its physical twin throughout its life-cycle, and that ultimately informs decisions that realize value”. Digital twins hold substantial potential in medicine, with key applications in cardiovascular disease, oncology, and surgery. They enable personalized treatments, uncover biological mechanisms, optimize surgical planning, predict prognosis, and integrate multimodal data. −

Analogously, in flow chemistry, digital twins provide a virtual counterpart of reactors and analytics by virtualizing physical entities in cyberspace, so software agents can exchange messages, actuate equipment, and coordinate experiments across geographically separated laboratories in real time. Kraft and co-workers implemented this idea within a dynamic knowledge graph architecture that uses ontologies and autonomous agents to drive goal directed experimentation workflows. They linked two automated flow platforms in Cambridge and Singapore to collaboratively optimize an aldol condensation reaction, targeting material cost and yield over variables including reagent equivalents, residence time, and reaction temperature.

Across 65 experiments, the real time collaboration accelerated progress on the Pareto front and achieved the highest yield of 93%.

Abolhasani et al. paired a reinforcement learning controller with a data trained digital twin to optimize multistep flow syntheses in closed loop. The reinforcement learning agent’s belief model consisted of an ensemble neural network regressor and a classifier. It used a rollout policy with an upper confidence bound decision rule to select the next injection action in real time. Abolhasani and co-workers trained a digital twin on prior campaigns to predict key outputs and injection viability. Then they used it as a stand in environment to compare optimization strategies without running excessive physical experiments. Drawing on this existing framework by Abolhasani and co-workers, it may be feasible to predict residence time distribution via the digital twin. Specifically, a digital twin trained on relevant data could first establish a mapping between operating conditions and residence time distribution representations.

The integration of digital twins and reinforcement learning also provides a practical and efficient pathway for automated solvent-switch optimization. Elmaz et al. presented an algorithmic framework for automated solvent-switch optimization using deep reinforcement learning. The process objective was to replace THF with 1-propanol to meet the preset final requirements while respecting safety limits such as minimum reboiler volume and maximum condenser load. A digital twin based on differential algebraic equations was used as a training and evaluation environment, enabling repeated policy improvement without excessive plant trials. The control policy was learned with proximal policy optimization, and operational constraints were embedded via logarithmic barrier functions to prevent constraint violations and keep the agent within safe operating regimes.

Rather than acting as a low-level controller, the reinforcement learning agent generated an optimized operating recipe that could be tracked by conventional control, providing a practical route to deploy AI-driven solvent-switch strategies.

2.5.2. Perspective on AI-Assisted Flow Synthesis

AI has been extensively utilized in drug target discovery, virtual screening, and novel drug design. We also anticipate its accelerated integration and advancement in synthetic route planning, process optimization, and continuous flow process development. Cloud-based laboratories will become increasingly prevalent, facilitating real-time data collection and remote experimentation. These platforms will allow for global collaboration and faster optimization of chemical processes. The influencing factors in continuous-flow reactions, particularly multistep continuous processes, are highly complex. These include the intrinsic kinetics of the reactions, the mixing characteristics within reaction chips, and the intricate interactions among multiple variables such as temperature, pressure, flow rate, and catalysts.

Researchers tend to be hesitant to publish ″unsuccessful″ or ″failed″ reaction data characterized by low yields or poor selectivity. This trend has fostered a somewhat overly optimiztic outlook within the chemical sciences where most reactions are assumed to proceed successfully. The scarcity of comprehensive data sets that include continuous flow synthesis reactions, especially negative ones, presents challenges for the advancement of AI-driven microflow synthesis. , Training AI model will rely not only on learning from existing literature data but also, and more critically, on high-throughput experimental platforms to generate authentic, reliable, and interconnected multidimensional databases. The integration of AI with automated flow chemistry will drive the next phase of innovation in flow chemistry, revolutionizing both research and industrial applications.

2.6. De novo Flow

The term “de novo” originates from Latin, meaning from the beginning. It describes approaches that operate without relying on prior templates, such as detecting protein or gene sequences without reference data, or AI-assisted drug design that does not depend on lead compounds. − In the context of process development, conventional methods that begin with batch reactions before transitioning to continuous flow are often inefficient. This has led to the emergence of the “de novo flow” approach, where the feasibility of continuous flow reactions should be initially assessed through batch experiments, with a focus on evaluating key physicochemical properties such as viscosity, solid content, and flow behavior, as these are critical for ensuring smooth operation in a continuous flow system. For flow process parameters like flow rate, temperature, concentration, pressure and reactant equivalents, the need for repeated screening through batch reactor trials can be bypassed. Instead, the screening and optimization of the aforementioned parameters are initiated directly in continuous flow reactors. By skipping batch-based development, “de novo flow” significantly enhances R&D efficiency. Traditional continuous process development, which heavily relies on empirical knowledge and manual effort, is increasingly seen as outdated. , The feature of “de novo flow” lies in data-driven strategies, like reaction kinetics, reactor geometry structure, and fluid dynamics, implemented by automated or AI-assisted instruments in “closed-loop” way. Noël and Jensen’s AI-assisted platforms are both typical cases of de novo flow. ,, Another example, Wu and co-workers developed an SPS-flow technology (solid-phase flow synthesis), particularly useful for de novo design and synthesis of drug molecules.

A typical de novo flow platform generally comprises microchannel reactor units, workup unit, pretreatment unit, online detection unit, and a program or AI control system. Since different online analytical instruments have varying sample requirements, the functions of the workup and pretreatment units must be specifically matched to the online detectors to ensure compatibility and accurate analysis.

To enable automated screening of reaction parameters, such as temperature, residence time, flow rate, concentration, and reactant ratios, the microreactor unit must be capable of automatically switching its operational modes based on instructions generated by program or AI. If the reactor module is separated from the platform, the remaining infrastructure essentially functions as an automated sample processing and analysis system. Through rational design and integration of its sample handling, pretreatment, and detection modules, such a platform can be adapted for the online analysis of samples from multiple sources, including foods, pharmaceuticals, synthetic compounds, plant extracts, or fermentation broths, etc.

3. Outlook and Conclusions

By focusing on the challenging aspects in multistep continuous synthesis, we systematically present relevant solutions reported in literature and our studies/practice, in hopes of providing guidance for researchers in this field. As described herein, multistep continuous synthesis represents a crucial approach for the future production of fine chemicals and serves as the process foundation for smart factories or unmanned “dark” plants. Its realization is highly challenging. It requires not only overcoming common difficulties encountered in single-step continuous reactions, such as gas participation or generation and the handling of solid-containing materials, but also addressing compatibility issues across multiple steps, including solvents, material, and residence time. To ensure stable multistep continuous synthesis, real-time online monitoring and tracking, as well as efficient and convenient workup and separation processes, are also critically important.

Also, the current developments and future trends in microflow technology are outlined, and perspectives on the future of the field are offered based on our understanding and practical experience. In particular, regarding research paradigms, the traditional approach of first developing batch reactions before transitioning to continuous processes will gradually be phased out. Instead, the data- and AI-driven “de novo flow” approach will become the trend in process development, relying heavily on automation and intelligent systems. The influencing factors in continuous-flow reactions, particularly multistep continuous processes, are highly complex. Whether “de novo flow” can function effectively in complex reaction systems remains an open challenge.

We believe that as the field advances, the community may gradually establish a cultural consensus on data sharing, and researchers may be more inclined to adopt standardized data formats to integrate various types of experimental data into AI-driven models. And a growing number of data repositories will be established and shared. This will help avoid the exploration pitfalls that researchers may have encountered previously, thereby advancing experimental research more efficiently. It is also foreseeable that, through innovations in reactors and separation methods, as well as deeper integration with electrochemistry and photochemistry, data- and AI-driven multistep continuous synthesis will progressively replace traditional synthetic methods. It is set to play an increasingly important role in the large-scale production of small molecules, peptides, nucleic acids, and more.

The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript. CRediT: Junjie Chen writing - original draft, writing - review & editing; Yibo Zou writing - review & editing; Lianan Liao writing - review & editing; Baochao Yang writing - review & editing; Dong Xing writing - review & editing; Jian Zhou writing - review & editing; Xuhong Qian writing - review & editing; Haibing He writing - original draft, writing - review & editing; Shuanhu Gao conceptualization, resources, supervision, writing - original draft, writing - review & editing.

This work is supported by the National Natural Science Foundation of China (22225105, 22301078, 22501083), National Key Research & Development Program of China (nos. 2022YFD1700402 and 2022YFD1700400) and “the Fundamental Research Funds for the Central Universities” and the Shanghai Frontiers Science Center of Molecule Intelligent Syntheses and the State Key Laboratory of Petroleum Molecular & Process Engineering (RIPP, SINOPEC, 2024KFR0042) and Huizhou Yinnong Technology Co., Ltd.

The authors declare no competing financial interest.

Published as part of JACS Au special issue “Continuous Flow Chemistry”.