Crystal Intentions: A Design-Integrated Roadmap for Engineering the Solid-State Properties of Active Pharmaceutical Ingredients

1. Introduction

The Solid Form as a Hidden Design Variable

Every crystal has an intention, in the sense that its final packing, hydration state, or lack of long-range order is the deterministic outcome of the molecular and process conditions under which it formed whether or not a formulation scientist chose those conditions deliberately. Left un-engineered, that intention is set by whatever the molecule happens to encounter first during discovery-stage crystallization, and is only as good, or as risky, as chance allows. Roughly two out of every five marketed oral drugs, and up to nine out of every ten candidates now entering discovery pipelines, are constrained by poor aqueous solubility and fall into Biopharmaceutics Classification System (BCS) Classes II and IV [1]. For these molecules, pharmacological potency is rarely the limiting factor; the limiting factor is whether the compound can be packaged into a solid that dissolves quickly enough, survives storage long enough, and can be manufactured reproducibly enough to reach a patient as an effective medicine [2]. That constraint is resolved, almost entirely, at the level of solid-state chemistry the three-dimensional arrangement, or absence of long-range arrangement, of molecules in the solid phase [3].

A single API molecule can, in principle, be isolated in dozens of distinct solid forms: multiple crystalline polymorphs, hydrates and organic solvate, ionic salts, non-ionic multicomponent cocrystals, co-amorphous binary systems, and fully amorphous glasses [4]. Each occupies a different point on the compound's configurational energy surface, with a distinct lattice energy, packing density, and surface chemistry, and each therefore differs in melting point, intrinsic solubility, dissolution rate, hygroscopicity, compressibility, and chemical reactivity despite, in the polymorphic case, being chemically identical down to the last atom [5]. The regulatory and commercial history of pharmaceuticals contains repeated cautionary examples in which a late-appearing or unanticipated solid-state transition altered clinical exposure, delayed approval, or forced a product withdrawal, and these episodes have converted solid-state form from a manufacturing footnote into a formally recognized critical quality attribute [6].

This review argues that the disparate strategies used to engineer solid-state form polymorph selection, salt and cocrystal design, amorphous dispersion, and particle engineering are more usefully understood not as five independent toolkits but as five ways of deliberately setting an API's crystal intentions by relocating it on a single underlying free-energy landscape. Section 2 develops this landscape framework explicitly [6]. Section 3 through Section 6 apply it in turn to polymorphism, multicomponent ionic and supramolecular engineering, amorphous and co-amorphous systems, and particle/habit engineering. Section 7 surveys the analytical instrumentation required to read the landscape experimentally, and Section 8 surveys the computational and machine-learning tools now used to predict it before a single crystal is grown. Section 9 considers sustainability and continuous-manufacturing constraints on landscape navigation, and Section 10 integrates all of the above into a single Solid-State-by-Design (SSbD) decision architecture a design-integrated roadmap for turning an API's crystal intentions from an accident of discovery chemistry into an engineered outcome. We draw throughout on primary literature published from 2020 onward, reflecting how substantially machine learning, crystal structure prediction, and continuous processing have reshaped this field within the current decade.

2. A Free-Energy-Landscape Framework for Solid-Form Selection

2.1. Two Orthogonal Axes: Thermodynamic Depth and Kinetic Accessibility

Every solid form of an API can be located, at least conceptually, on a lattice (or, for multicomponent and amorphous systems, a configurational) free-energy surface, G, plotted against a generalized packing coordinate that captures molecular conformation, orientation, and intermolecular contact geometry [7]. Figure 1 renders this surface schematically for a hypothetical API: the global minimum corresponds to the thermodynamically most stable polymorph (Form I); a shallower local minimum corresponds to a kinetically accessible but thermodynamically metastable polymorph (Form II); a third well, shifted by incorporation of water into the lattice, represents a hydrate; and a broad, shallow depression at high free energy represents the amorphous state a kinetically arrested, glassy configuration rather than a true energetic minimum [8].

Two properties of this surface are engineeringly decisive. The first is thermodynamic depth: how far below competing forms a given minimum sit, which determines how strongly the system resists reverting to a more stable form during storage, formulation, or exposure to the gastrointestinal environment. The second is kinetic accessibility: how readily, and how reproducibly, a given minimum can be reached from solution or melt under manufacturable conditions, and how high the barrier is to escape it once formed [9]. A deep, easily accessible minimum the ideal case is rare; in practice, every solid-state strategy discussed in this review represents a different negotiated trade-off between these two properties.

2.2. Reframing Five Engineering Levers in Landscape Terms

Table 1 restates the five principal solid-state engineering levers in landscape language, and Figure 2 plots them schematically along the two axes just defined. Polymorph and hydrate control seeks the deepest, most kinetically robust minimum reachable by the unmodified molecule and is therefore positioned toward the high-stability, high-accessibility quadrant, constrained only by the risk of disappearing polymorphism the discovery, sometimes years after initial development, of an even more stable form that was kinetically inaccessible during original screening [10]. Salt formation and co-crystallization do not merely search the native landscape more thoroughly; they construct an entirely new multicomponent landscape by introducing a counterion or coformer, trading some manufacturing complexity for large, often predictable gains in solubility. Amorphous and co-amorphous engineering deliberately abandons the deep crystalline minima altogether, accepting a thermodynamically elevated, kinetically metastable state in exchange for the largest attainable dissolution advantage a strategy that only succeeds when a stabilizing carrier raises the effective kinetic barrier to recrystallization high enough to survive shelf life. Particle and crystal habit engineering operates on an essentially independent, mesoscale landscape: it does not change which molecular-level minimum is occupied, but reshapes the particle-level energy surface governing agglomeration, flow, and surface area, and can therefore be layered on top of any of the other four strategies.

3. Polymorphism and Pseudopolymorphism: Navigating Local Minima

3.1. Origins of Polymorphic Diversity

Polymorphs are distinct crystalline packings of an identical molecule, each corresponding to a local or global minimum on the native lattice free-energy surface. Because the strength and geometry of intermolecular contacts differ between packings, polymorphs differ measurably in lattice energy and therefore in melting point, apparent solubility, dissolution rate, and mechanical behavior during compaction, even though no covalent bond has changed [11,12]. A metastable polymorph one occupying a shallower well than the global minimum typically dissolves faster and shows higher apparent solubility, which is attractive from a bioavailability standpoint, but carries the risk of solution-mediated or solid-state conversion to the more stable form during storage, formulation processing, or transit through the gastrointestinal tract, at which point the kinetic advantage is lost [13].

3.2. Disappearing Polymorphism and the Limits of Screening

Modern polymorph screening combines high-throughput parallel crystallization across many solvent, temperature, and supersaturation conditions with slurry-based competitive equilibration experiments intended to identify the thermodynamically stable form under defined storage conditions. Seeding strategies, programmed cooling profiles, and habit-modifying additives are standard tools for steering crystallization toward a specific target polymorph at manufacturing scale [14,15]. However, screening alone cannot guarantee completeness: the well-documented phenomenon of disappearing polymorphism describes cases in which a previously unobserved, more stable form appears sometimes irreversibly displacing the originally marketed form only after years of manufacturing history, as occurred famously with ritonavir [16]. Because experimental screening can never formally prove the absence of an undiscovered deeper minimum, regulatory frameworks increasingly require that a polymorph control strategy be engineered directly into the manufacturing process itself, rather than relying on end-product testing alone [17,18].

3.3. Hydrates, Solvates, and Humidity-Driven Transitions

Hydrates and solvates collectively, pseudo-polymorphs add a further dimension to the native landscape by incorporating solvent molecules, most often water, into the crystal lattice in stoichiometric or non-stoichiometric ratios [19,20]. Because hydrate formation can occur spontaneously under ordinary ambient humidity, an anhydrous form selected during early-phase development on the basis of superior solubility may convert unintentionally during granulation, coating, packaging, or long-term storage, silently eroding the intended dissolution advantage [20,21]. Dynamic vapor sorption and humidity-controlled powder X-ray diffraction remain the principal experimental tools for mapping this humidity-dependent region of the landscape, and are discussed further in Section 7 [22].

4. Ionic and Supramolecular Multicomponent Engineering

4.1. Salts: Shifting the Ionization Equilibrium

Salt formation remains the single most widely used strategy for improving the solubility and dissolution of ionizable APIs, underlying roughly half of all marketed small-molecule medicines [23]. By pairing an acidic or basic drug with a pharmaceutically acceptable counterion, developers construct an entirely new ionic crystal lattice whose solubility, hygroscopicity, and melting behavior can differ dramatically from the free acid or base, even though the pharmacologically active species is regenerated upon dissolution [24]. Rational salt selection begins with the counterion pKa relative to the API a differential of at least roughly two pH units is generally required to favor complete, energetically stable proton transfer over a poorly defined ionic complex and continues through assessment of the resulting salt's crystallinity, hygroscopicity, and counterion toxicity.

4.2. Disproportionation as a Landscape-Crossing Event

A pharmaceutical salt can be understood as sitting in a favorable well on an ionic landscape only so long as local pH and counterion volatility conditions keep that well energetically preferred over reversion to the free acid or base [25]. Disproportionation the in-situ dissociation of a salt back to its neutral form, typically triggered by the alkaline microenvironment of certain excipients or by loss of a volatile counterion such as hydrochloride is therefore best understood as an unwanted crossing back over the barrier separating the salt well from the free-form well [26]. Recent mechanistic and modeling work has quantified this risk directly: sulfonate counterions have been shown to modulate biopharmaceutical performance across the population of FDA-approved salts in structurally systematic ways, refined pHmax equations now allow in-silico prediction of the pH at which a given salt becomes thermodynamically disfavored relative to its free form, and advanced synthetic methodologies have expanded the practical space of counterions available for salt screening beyond the classical shortlist [27,28]. These tools collectively let formulators identify, before a single stability study is run, which candidate salts sit in genuinely deep wells versus which are only kinetically, and precariously, trapped [28,29].

4.3. Cocrystals: Synthon-Directed Lattice Redesign

For non-ionizable APIs, or where the counterion-toxicity or disproportionation risk of a salt is unacceptable, pharmaceutical cocrystals offer a complementary multicomponent route. A cocrystal is a crystalline solid in which an API and a pharmaceutically acceptable coformer coexist in a defined stoichiometric ratio within a single lattice, held together by non-covalent interactions hydrogen bonding, halogen bonding, and π–π stacking without proton transfer [30]. Because cocrystallization leaves the covalent structure of the API untouched, it decouples physicochemical property tuning from pharmacological identity almost entirely, and comprehensive 2022 and subsequent reviews of cocrystal engineering document how thoroughly this platform has matured from serendipitous discovery toward literature-scale, database-mined design.

Coformer selection is now guided systematically by supramolecular synthon theory, which predicts the most probable hydrogen-bonding or halogen-bonding motifs between complementary functional groups on the API and candidate coformer, together with computational hydrogen-bond-propensity and Cambridge Structural Database mining tools that rank candidate coformers by predicted cocrystallization likelihood before any experimental screening begins. This computational-first approach has been extended specifically to low-molecular-weight and flavonoid-class APIs, and machine-learning-assisted coformer and multicomponent-form prediction pipelines are increasingly used to prioritize the coformer shortlist for experimental confirmation, converting what was once a combinatorial search problem into a ranked, hypothesis-driven screening campaign.

4.4. Regulatory Boundary Between Salts, Cocrystals, and Polymorphs

Because cocrystals occupy conceptual territory between a polymorph (no new chemical species) and a salt (complete proton transfer), regulatory agencies have published dedicated frameworks distinguishing cocrystals from both categories for the purposes of classification, patenting, and lifecycle management. This regulatory clarity has removed a significant historical barrier to cocrystal adoption in commercial development, and well-characterized novel multicomponent forms whether salts or cocrystals now routinely extend intellectual property protection for an API well beyond the life of the original composition-of-matter patent.

Table 2. Comparative attributes of pharmaceutical salts and cocrystals as ionic versus neutral multicomponent landscape strategies.

Attribute	Pharmaceutical Salts	Pharmaceutical Cocrystals
Requires ionizable functional group	Yes (ΔpKa ≥ ~2 typical threshold)	No, applicable to neutral APIs
Bonding mechanism	Ionic (proton transfer)	Non-covalent (H-bond, halogen bond, π–π)
Principal instability mode	Disproportionation to free acid/base	Dissociation to component crystals under humidity/solvent stress
Screening logic	Counterion pKa matching; pHmax modeling	Supramolecular synthon prediction; CSD mining; ML ranking
Regulatory classification	Distinct new drug substance if patentable salt	Clarified as distinct from polymorphs/salts by dedicated guidance
Typical solubility gain	Large and comparatively predictable	Moderate to large; more coformer-dependent

Table 2. Comparative attributes of pharmaceutical salts and cocrystals as ionic versus neutral multicomponent landscape strategies.

Attribute	Pharmaceutical Salts	Pharmaceutical Cocrystals
Requires ionizable functional group	Yes (ΔpKa ≥ ~2 typical threshold)	No, applicable to neutral APIs
Bonding mechanism	Ionic (proton transfer)	Non-covalent (H-bond, halogen bond, π–π)
Principal instability mode	Disproportionation to free acid/base	Dissociation to component crystals under humidity/solvent stress
Screening logic	Counterion pKa matching; pHmax modeling	Supramolecular synthon prediction; CSD mining; ML ranking
Regulatory classification	Distinct new drug substance if patentable salt	Clarified as distinct from polymorphs/salts by dedicated guidance
Typical solubility gain	Large and comparatively predictable	Moderate to large; more coformer-dependent

5. Amorphous and Co-Amorphous Engineering: Trading Depth for Height

5.1. The Spring-and-Parachute Mechanism

Where crystalline strategies reach their limit typically for high-melting, poorly ionizable, hydrogen-bond-poor molecules that resist both salt and cocrystal formation amorphous solid dispersions (ASDs) offer a fundamentally different route to enhanced apparent solubility. In an ASD, the API is molecularly dispersed within a polymeric carrier, eliminating the lattice energy barrier that otherwise limits dissolution from the crystalline state and permitting dissolution to generate a transiently supersaturated solution the spring whose degree of supersaturation can exceed thermodynamic equilibrium solubility by an order of magnitude or more [31]. Figure 3 depicts this behavior schematically: without a stabilizing polymer, the supersaturated spring collapses rapidly as the API recrystallizes from solution, erasing the advantage within hours; with an appropriately selected polymer, dissolved polymer chains inhibit nucleation and crystal growth in solution, sustaining supersaturation long enough for absorption to occur the parachute.

5.2. Polymer Selection as Landscape Stabilization

Because the amorphous state is thermodynamically metastable relative to every crystalline form on the native landscape, the central engineering challenge of ASD design is not reaching the amorphous state but keeping the API there for the duration of shelf life. The stabilizing polymer must satisfy at least three simultaneous requirements: it must kinetically inhibit nucleation and crystal growth during storage, it must remain miscible with the API across the intended shelf-life temperature and humidity range (a requirement often assessed through solubility-parameter matching and Flory–Huggins-type miscibility modeling), and it must itself inhibit recrystallization from the supersaturated solution generated on dissolution. Recent large-dataset machine-learning studies including models trained on hundreds of hot-melt-extruded formulations spanning dozens of distinct APIs have shown that gradient-boosted and deep-learning classifiers can now predict both amorphization success and post-manufacture chemical stability directly from molecular fingerprints and polymer descriptors with accuracies exceeding 90%, substantially compressing what was historically a purely empirical, trial-and-error polymer-selection process [32]. Complementary work has extended this predictive capability to drug-loading limits in poor glass-forming APIs and to decision-tree-style formulation guidance for super saturable mesoporous silica carriers, broadening the ASD toolkit beyond conventional polymer dispersions.

5.3. Co-Amorphous Systems as a Low-Molecular-Weight Alternative

Co-amorphous systems (CAMs) replace the high-molecular-weight polymeric carrier with a second small molecule either a genuine second drug (drug–drug CAMs) or a low-molecular-weight coformer such as an amino acid or carboxylic acid forming a single-phase amorphous blend stabilized by intermolecular hydrogen bonding, ionic interaction, or π–π stacking rather than by polymer entanglement. Because both components are small molecules, CAMs can achieve markedly higher glass transition temperatures and drug loadings than conventional polymeric ASDs, and predictive coformer-selection frameworks increasingly informed by computational screening of intermolecular interaction strength have begun to convert CAM design from a case-by-case empirical exercise into a more systematic screening campaign, mirroring the trajectory already established for cocrystal coformer selection.

5.4. Manufacturing Routes: HME Versus Spray Drying

Two manufacturing platforms dominate commercial ASD production: spray drying, which rapidly removes solvent from an atomized API–polymer solution, and hot-melt extrusion (HME), a solvent-free continuous process in which thermal and mechanical energy are applied to a molten API–polymer blend. Each imposes distinct constraints spray drying requires a common solvent system and generates fine, often low-bulk-density powders that need downstream densification, while HME requires thermal stability of the API at processing temperature but offers a genuinely continuous, solvent-free, and therefore comparatively green manufacturing route consistent with current regulatory encouragement of continuous processing. Systematic reviews of HPMCAS-based ASDs now marketed clinically confirm that both platforms are represented among commercially approved products, and quality-by-design frameworks specifically developed for HME-based ASD product development formalize design-of-experiments approaches to screening formulation and process variables jointly rather than sequentially [33].

6. Particle and Crystal Habit Engineering: Coupling Molecular Form to Powder Performance

Molecular-level solid-state form determines only part of a formulation's ultimate performance; particle-level properties size, shape, surface area, and habit govern how that molecular form actually behaves during downstream unit operations and dissolution. The Noyes–Whitney relationship makes this coupling explicit: dissolution rate scales directly with the exposed surface area of the solid, so even a thermodynamically well-chosen polymorph, salt, or cocrystal can underperform if its particle morphology yields poor powder flow, inconsistent tablet compaction, or insufficient surface area for rapid dissolution.

6.1. Habit Control During Crystallization

Crystal habit the relative growth rates of different crystal faces, and therefore the overall particle shape is influenced by solvent choice, supersaturation level, cooling profile, and habit-modifying additives that adsorb selectively onto specific crystal faces and slow their growth relative to others. Controlled crystallization platforms, including antisolvent precipitation, programmed cooling profiles, and continuous mixed-suspension mixed-product-removal (MSMPR) crystallizers, increasingly allow particle size distribution and habit to be engineered concurrently with polymorphic form selection at the crystallization step itself, rather than corrected after the fact by milling a post hoc process that can introduce surface disorder, amorphous content, and physical instability at particle surfaces [34].

6.2. Spherical Crystallization and Co-Processing

Spherical crystallization and continuous spherical agglomeration produce near-spherical, free-flowing, directly compressible particles from the crystallization step itself, illustrating how particle engineering and molecular crystal engineering are converging into a single upstream design activity. An integrated continuous crystallization–spherical agglomeration (CCSA) process demonstrated for atorvastatin calcium exemplifies this convergence, coupling continuous polymorph control with in-line particle shaping in a single intensified unit operation, while systematic studies of bridging-liquid selection and two-step bridging mechanisms have improved the reproducibility of agglomerate size control for benzoic-acid-class and arbidol-hydrochloride-class model APIs. A comprehensive 2023 review of API co-processing techniques catalogs spherical agglomeration and co-precipitation as complementary, solvent-media-based approaches that simultaneously address flow, compaction, and dissolution limitations without requiring a separate granulation step.

6.3. Nanocrystals as an Orthogonal Particle-Size Lever

At the opposite end of the particle-size spectrum, nanocrystal engineering reduces API particles to the sub-micron domain, exploiting the direct proportionality between surface area and dissolution rate to achieve solubility and bioavailability gains without altering molecular solid-state form at all. Recent formulation work has demonstrated fully redispersible dried nanocrystal powders stabilized with sucrose laurate that retain enhanced surface area and dissolution advantage after reconstitution, and combined co-crystal/nanocrystal approaches in which a cocrystal is itself reduced to nanoscale particle size by wet milling illustrate that molecular-level and particle-level engineering are not mutually exclusive but can be stacked to compound their individual solubility benefits, consistent with the particle-engineering axis in Figure 2 operating independently of, and in combination with, the molecular-landscape axis [35].

7. The Analytical Backbone: Reading the Landscape Experimentally

Every claim this review has made about landscape position which minimum an API occupies, how deep it is, and whether it is drifting toward another well is only as credible as the analytical methods used to verify it. Table 3 summarizes the core techniques required to detect and quantify solid-state heterogeneity, often at low levels within a bulk sample, and no single technique is sufficient in isolation: robust characterization depends on triangulating orthogonal methods, particularly when quantifying trace levels of a minor polymorph or residual crystallinity within a nominally amorphous matrix, since individual techniques carry different and sometimes complementary limits of detection [36].

Solid-state NMR in particular has undergone substantial methodological advance since 2020, with rapid ¹H quantification protocols and dynamic-nuclear-polarization-enhanced detection now extending the sensitivity of ssNMR into the sub-percent regime for polymorph and amorphous-content quantification a resolution that PXRD alone often cannot reliably achieve for low-level minor phases [37]. These advances matter disproportionately for the landscape framework proposed here, because the practical difference between a genuinely deep, safe minimum and a shallow, drifting one is frequently a matter of a percent or two of a minor competing phase that only the most sensitive orthogonal techniques can detect before it becomes a clinical or manufacturing problem.

8. Computational and AI-Accelerated Screening

8.1. Crystal Structure Prediction Matures into a Design Tool

Perhaps the most transformative development in solid-state pharmaceutics over the past five years is the maturation of crystal structure prediction (CSP) from an academic curiosity into a routinely deployable industrial tool. CSP generates and energy-ranks plausible crystal packing arrangements for a candidate molecule computationally, before any experimental crystallization is attempted, effectively mapping the theoretical free-energy landscape in silico ahead of the experimental search depicted in Figure 1. Recent validation studies combining systematic crystal-packing search algorithms with machine-learned interatomic force fields have demonstrated recovery of essentially all experimentally known polymorphic forms across large, diverse validation sets spanning dozens of molecules and well over a hundred confirmed polymorphs, while simultaneously flagging additional low-energy structures not yet observed experimentally precisely the disappearing-polymorphism risk discussed in Section 3.2, now surfaced computationally rather than discovered by costly post-launch reformulation. Complementary work applying active-learning-trained machine-learned potentials to CSP energy landscapes has improved the accuracy with which candidate structures are ranked relative to the prohibitively expensive periodic density-functional-theory calculations that CSP traditionally required, while broader field reviews confirm that CSP is now sufficiently reliable to inform polymorph screening strategy directly and to support prediction of physical stability under real-world storage conditions rather than only idealized ones.

8.2. Machine Learning Across the Pipeline

Machine learning has been layered onto essentially every stage of the solid-state design pipeline discussed in this review. Beyond CSP itself, ML models trained on historical crystallization and cocrystallization datasets now predict cocrystallization propensity and coformer suitability from molecular descriptors ahead of experimental synthon screening, ML classifiers trained on hundreds of hot-melt-extruded formulations predict amorphization success and post-manufacture chemical stability of ASDs with better than 90% accuracy, and combined high-throughput-screening/ML pipelines predict binary and ternary ASD formation from micro-quantity experimental datasets an order of magnitude faster than exhaustive combinatorial screening alone [38]. None of these tools yet replace experimental verification every study reviewed here retains a confirmatory experimental step but collectively they compress design cycles by focusing laboratory effort on the small subset of candidate forms that computational screening identifies as most promising, converting solid-state form selection from a largely serendipitous search into a ranked, hypothesis-driven engineering exercise.

9. Sustainability and Continuous Manufacturing as Landscape Constraints

Solvent use, energy demand, and waste generation associated with crystallization are increasingly treated not as afterthoughts but as explicit design constraints on which region of the solid-state landscape is practically reachable at manufacturing scale. Continuous crystallization platforms most commonly built around mixed-suspension mixed-product-removal (MSMPR) crystallizer configurations offer enhanced process robustness, tighter control over polymorphic outcome, and substantial reductions in solvent burden relative to traditional batch crystallization, and have received explicit regulatory encouragement through frameworks such as ICH Q13, which formally addresses continuous manufacturing of both drug substances and drug products [39]. Recent technoeconomic and sustainability analyses comparing batch and continuous crystallization workflows using open-source pharmaceutical flowsheet simulation tools now allow developers to quantify these tradeoffs quantitatively before committing to a manufacturing platform, while green-chemistry-oriented reviews document a measurable, if still limited, convergence between continuous manufacturing adoption and green-chemistry principle implementation across the pharmaceutical and pharma-chemical industries since roughly 2020. Deep-eutectic-solvent-based crystallization media represent a further emerging green-chemistry lever, offering tunable solubility and polymorphic control while substantially reducing reliance on conventional volatile organic solvents during the crystallization step itself.

10. Toward an Integrated Solid-State-by-Design (SSbD) Architecture

Figure 4 translates the landscape framework of Section 2 into an operational decision tree that mirrors typical early-development triage logic: ionizability is assessed first, favoring salt formation when a sufficient pKa differential is available; cocrystallization is evaluated next for non-ionizable APIs or where disproportionation risk disfavors a salt; and amorphous or co-amorphous engineering is reserved for molecules that remain intractable high-melting, poorly crystallizable, or lacking a suitable coformer after crystalline strategies have been exhausted. Polymorph and hydrate screening runs as a parallel, always-on activity across every branch rather than as a sequential step, consistent with the regulatory expectation, discussed in Section 3.2 and Section 7, that polymorphic risk be controlled continuously rather than assessed once. Particle and crystal habit engineering is applied concurrently to whichever crystalline or amorphous form is ultimately selected, reflecting its status as an independent, mesoscale landscape rather than a downstream afterthought.

Taken together, the strategies reviewed in Section 3 through Section 8 are most powerful when integrated early and iteratively, rather than applied sequentially as isolated late-stage fixes. We propose treating solid-state engineering analogously to Quality by Design (QbD) in drug product manufacturing: critical material attributes of the solid form crystallinity, particle size and shape, surface energy, hygroscopicity should be identified early, linked explicitly to critical quality attributes of the final product dissolution rate, bioavailability, physical and chemical stability, manufacturability and controlled through a deliberately engineered combination of molecular design, computational screening, crystallization or amorphization process parameters, and downstream particle processing. Figure 5 depicts this Solid-State-by-Design (SSbD) logic as a closed decision cycle, and Table 4 maps each stage of that cycle to the critical material attributes it is responsible for controlling and the analytical methods (Section 7) used to verify that control.

Open Challenges and Future Perspectives

• Long-term amorphous stability prediction remains difficult: accelerated stability protocols do not always predict shelf-life recrystallization behavior reliably, and while machine-learning models trained on hundreds of formulations now predict short-term amorphization and manufacture-stage chemical stability with high accuracy, prediction of multi-year physical stability under real-world, variable storage conditions remains comparatively immature.
• Extension of CSP and ML toolkits to larger, more conformationally flexible molecules: current CSP validation successes concentrate on comparatively rigid, historically well-studied small molecules; as discovery pipelines increasingly nominate larger, more flexible, more poorly soluble candidates, existing crystal-energy-landscape search algorithms face an exponential growth in conformational degrees of freedom that current machine-learned potentials only partially mitigate.
• Deeper integration of green chemistry and continuous manufacturing: despite explicit regulatory encouragement through ICH Q13 and a growing continuous-crystallization literature base, adoption of fully continuous, solvent-minimized crystallization at commercial scale remains limited relative to its demonstrated technoeconomic and sustainability advantages, and deep-eutectic-solvent and other green-solvent crystallization media remain at a comparatively early stage of pharmaceutical-scale validation.
• Earlier integration of computational screening into candidate selection itself: CSP, ML-guided coformer ranking, and ASD-formation prediction are currently deployed almost exclusively after a lead molecule has already been nominated; closer integration of these tools directly into early discovery decision-making could allow predicted solid-state developability to influence which candidate is advanced in the first place, rather than only how an already-selected candidate is formulated.
• Systematic coformer and co-amorphous-partner databases: despite substantial recent progress in ML-guided coformer ranking for both cocrystals and co-amorphous systems, the underlying training datasets remain comparatively small and unevenly distributed across chemical space, constraining the generalizability of current predictive models to structurally novel APIs.

11. Conclusions

Solid-state form is not incidental to drug development, nor separable from it; it is a design variable with the same strategic weight as molecular structure and dosage-form selection. Reframing polymorph control, salt and cocrystal engineering, amorphous and co-amorphous dispersion, and particle engineering as five coordinated ways of navigating a single underlying free-energy landscape rather than as five disconnected toolkits clarifies why each strategy succeeds or fails where it does, and provides a unifying vocabulary for comparing gains in thermodynamic depth against gains in kinetic accessibility. The rapid maturation since 2020 of crystal structure prediction using machine-learned interatomic potentials, machine-learning-guided coformer and amorphous-dispersion screening, quantitative disproportionation risk modeling, and continuous, solvent-minimized crystallization platforms has, for the first time, made much of this landscape computationally navigable before a single experimental crystal is grown. Realizing the full benefit of these tools requires treating solid-state form selection as an integrated Solid-State-by-Design activity from the earliest stages of development the framework summarized in Figure 1 through Figure 5 and Table 1 through 4 of this review rather than as a sequence of isolated fixes applied late, and often too late, in the development timeline.