Patent drawings are a core component of patent applications — not supplementary illustrations of the text, but the visual anchors of the protection scope. Drawing reference numerals correspond directly to technical features in the claims; drawing types directly reflect the thinking patterns of the technical domain. A patent application that lacks correct drawings, however well written, may be rejected for "insufficient disclosure in the specification."

This article analyzes the drawing requirements of seven technical domains and explains how to select between Mermaid and AI image generation technology to produce patent drawings that comply with the format standards of each major jurisdiction.

Why Drawing Requirements Differ by Technical Domain

Different technical domains use drawings to convey fundamentally different types of information. This is not a stylistic preference — it is a content-driven necessity:

Software/communications patents convey execution sequences and system structures — the boxes and arrows of a flowchart correspond to steps in method claims; the modules in an architecture diagram correspond to functional units in apparatus claims. Without flowcharts, method steps lack intuitive illustration; without architecture diagrams, module connection relationships are difficult to understand.

Mechanical patents convey spatial and connection relationships — cross-sections show internal structures invisible to the naked eye; exploded views show the assembly relationships between components; detail views show close-up connections. These drawings are "the engineer's language" and are the foundation of sufficient disclosure in mechanical patents.

Chemical/pharmaceutical patents convey compound structures and experimental results — chemical structural formulas show the atomic connectivity of compounds; spectra (NMR, IR, XRD) serve as product characterization; experimental data curves show technical effects. Some purely formulation-based chemical patents can have no drawings at all.

Medical device patents combine mechanical and software drawing needs — 3D views and cross-sections show device structure; operation flowcharts show usage steps; system block diagrams show the electronic control components.

Optical/semiconductor patents have unique drawing types — layered cross-sections show each layer in a semiconductor device; band diagrams show electronic energy levels; optical path diagrams show light propagation — drawing types rare or nonexistent in other domains.

Civil/construction patents use engineering drawing conventions — floor plans, elevations, cross-sections, and joint details are the standard expressions of architectural engineering, with different drafting standards and emphases from mechanical cross-sections.

The Seven-Domain Drawing Classification System

CNIPA.AI organizes patents into 7 major technical categories, each with a complete drawing requirements specification:

1. Mechanical Structure (MECHANICAL)

Typical drawing count: 4–8. IPC: Class B, F01–F04, F15–F17.

Drawing Type	Generation Method	Required?	Typical Title Example
3D/perspective view	AI image generation	Mandatory	Fig. 1 Overall perspective diagram
Cross-section/sectional view	AI image generation	Mandatory	Fig. 2 A-A sectional view
Exploded view	AI image generation	Recommended	Fig. 3 Component exploded diagram
Assembly structure diagram	AI image generation	Recommended	Fig. 4 Assembly relationship diagram
Functional module block diagram	Mermaid	Optional	Fig. 5 System functional module diagram
Working flow diagram	Mermaid	Optional	Fig. 6 Working process diagram

The core mechanical drawings are the 3D perspective view and the cross-section — the perspective view gives examiners and judges an overall impression; the cross-section showing internal structure is the foundation of sufficient disclosure. The cutting position and direction of cross-sections must be clearly stated in the specification ("cross-sectional view along direction A-A"), and section reference numerals must be completely consistent with the body text.

2. Electronic/Electrical (ELECTRONIC_ELECTRICAL)

Typical drawing count: 3–6. IPC: H01, H02, H03, H05.

Drawing Type	Generation Method	Required?
Circuit schematic	AI image generation	Mandatory
System block diagram	Mermaid	Mandatory
Device cross-section	AI image generation	Recommended
Signal flow diagram	AI image generation	Recommended
Signal waveform diagram	AI image generation	As needed

Circuit schematics must use standard circuit symbols (resistors, capacitors, diodes, op-amps, etc.), with component designators (R1, C1, D1, etc.) consistent with the specification. System block diagrams are generated using Mermaid to show module connection relationships.

3. Software/Telecommunications (SOFTWARE_TELECOM)

Typical drawing count: 5–10. IPC: G06, G16, H04.

Drawing Type	Generation Method	Required?	Typical Title Example
Method flowchart	Mermaid	Mandatory	Fig. 1 Method overall flowchart
System architecture diagram	Mermaid	Mandatory	Fig. 2 System architecture diagram
Sequence/interaction diagram	Mermaid	Recommended	Fig. 3 Module interaction sequence diagram
Data flow block diagram	Mermaid	Recommended	Fig. 4 Data processing flow diagram
State transition diagram	Mermaid	As needed	Fig. 5 State transition diagram
Data flow diagram	Mermaid	As needed	Fig. 6 Data flow diagram
Class/module structure diagram	Mermaid	As needed	Fig. 7 Core class structure diagram
User interface mockup	AI image generation	As needed	Fig. 8 User interface mockup

Software/telecommunications is the category requiring the most drawings among the seven domains, and the vast majority can be generated using Mermaid. Structural accuracy (step sequence, diamond-shaped decision branches, loop arrows) matters more than visual appearance.

4. Optics/Semiconductors (OPTICS_SEMICONDUCTOR)

Typical drawing count: 4–8. IPC: G02, H01L, H10, B32B.

Drawing Type	Generation Method	Required?
Layered structural cross-section	AI image generation	Mandatory
Process flowchart	Mermaid	Mandatory
Optical path/wave diagram	AI image generation	Recommended
Package structure diagram	AI image generation	As needed
Band diagram	AI image generation	As needed
Current-voltage characteristic curve	AI image generation	As needed

Optics/semiconductor patents have a unique "layered structural cross-section" requirement — showing each thin-film layer in a semiconductor device (substrate, epitaxial layer, gate dielectric, electrode layers, etc.) with each layer annotated by material name or number, and spatial relationships (stacked/adjacent/embedded) clearly expressed. Band diagrams are important analytical tools in semiconductor physics and almost never appear in other technical domains.

5. Chemical/Pharmaceutical (CHEMICAL_PHARMA)

Typical drawing count: 0–3 (note the minimum of 0 — the only category in the seven that allows no drawings). IPC: A61K, A61P, A61Q, Class C, Class D.

Drawing Type	Generation Method	Required?
Preparation/synthesis flowchart	Mermaid	As needed
Compound structural formula	Specialized chemical rendering	As needed (mandatory for compound-type)
Reaction scheme	AI image generation	As needed
Metabolic/pathway diagram	AI image generation	As needed
Experimental data curve	AI image generation	As needed

The drawing strategy for chemical/pharmaceutical patents is fundamentally different from other domains. Pure formulation patents (composition claims specifying component concentration ranges) typically require no drawings; patents involving specific compounds require chemical structural formulas (generated by specialized chemical rendering tools, not Mermaid or AI images); experimental data curves (dose-response relationships, efficacy comparison curves), when presented graphically in the text, must be submitted as drawings.

Chemistry is the only patent domain in which photographs are allowed as drawings (e.g., micrographs showing metallographic structures or tissue cells) — an explicit provision in the CNIPA Patent Examination Guidelines.

6. Medical Devices (MEDICAL_DEVICE)

Typical drawing count: 4–7. IPC: A61B, A61C, A61F, A61M.

Drawing Type	Generation Method	Required?
Overall device 3D view	AI image generation	Mandatory
Key section cross-section	AI image generation	Mandatory
Assembly exploded view	AI image generation	Recommended
Use/operation flowchart	Mermaid	Recommended
System component block diagram	Mermaid	Recommended

Medical devices combine mechanical and software drawing needs — the device body uses AI image generation (3D views, cross-sections); electronic control systems and operation flows use Mermaid. A special requirement for medical device drawings is showing the device's relationship with the human body (e.g., implant position), which demands high accuracy.

7. Civil/Construction (CIVIL_CONSTRUCTION)

Typical drawing count: 3–6. IPC: Class E.

Drawing Type	Generation Method	Required?
Structural overall 3D view	AI image generation	Mandatory
Cross-sectional view	AI image generation	Mandatory
Architectural floor plan	AI image generation	Recommended
Architectural elevation	AI image generation	Recommended
Key connection joint detail	AI image generation	Recommended
Construction process flowchart	Mermaid	Recommended

Civil/construction patents use engineering drawing language; the drafting conventions for floor plans, elevations, cross-sections, and details differ significantly from mechanical drawings (scale, section annotation methods, material fill symbols). Construction method patents (method claims) require construction process flowcharts generated using Mermaid.

Mermaid vs. AI Image Generation: Decision Rules

This is the most important technical choice in patent drawing generation. The rule is clear:

Use Mermaid — when the core value of the drawing lies in the accuracy of structural relationships: step execution order (flowcharts), module connection relationships (architecture diagrams/block diagrams), message timing (sequence diagrams), state transitions (state diagrams). Mermaid takes text descriptions as input and generates standardized graphics through a rendering engine, guaranteeing that step sequences are not scrambled, decision branch conditions are logically correct, and arrow directions are accurate.

Use AI image generation — when the core value of the drawing lies in the visual rendering of physical form: the three-dimensional shape of mechanical components (3D views), the cross-sectional form of internal structures (sectional views), the symbols and connections of circuit elements (circuit diagrams), the layered structure of devices (semiconductor cross-sections). These drawings need visual credibility that Mermaid cannot express.

Use specialized chemical rendering — chemical structural formulas (benzene rings, functional groups, bond angles) require dedicated chemical structure rendering tools (such as RDKit or ChemDraw-format SVG exports); neither Mermaid nor AI images are suitable.

Drawing Type	Recommended Generation Method	Reason
Method flowchart	Mermaid	Step sequence accuracy is paramount
System architecture/block diagram	Mermaid	Module connection relationship accuracy is paramount
Sequence/state diagram	Mermaid	Logical relationship accuracy is paramount
Mechanical 3D view/cross-section	AI image generation	3D form visual rendering
Circuit schematic	AI image generation	Standard circuit symbol expression
Semiconductor cross-section	AI image generation	Precise expression of layer structure
Chemical structural formula	Specialized chemical rendering	Bond angles and atomic annotation
UI interface mockup	AI image generation	Interface appearance visual rendering

Jurisdiction-Specific Drawing Format Requirements: CN/US/JP/KR Differences

Drawing content requirements vary by technical domain; drawing format requirements vary by jurisdiction. Here are the key format differences among major jurisdictions:

China (CN) Drawing Format

CNIPA requires black-and-white line drawings (no color; photographs are an exception); drawing number format is "图1," "图2"; component reference numerals use Arabic digits ("1," "2," "3"); drawing captions in the specification read "Fig. 1 is a schematic diagram of..."; drawings may not contain detailed textual explanations — only reference numerals and necessary technical terms are permitted.

United States (US) Drawing Format

USPTO requires black-and-white line drawings (color must be specifically requested); drawing number format is "FIG. 1," "FIG. 2"; specification references to drawings write "FIG. 1" (with period); reference numerals are typically Arabic digits, with sub-numerals using letters (e.g., "12a," "12b"); 37 CFR 1.84 specifies paper size (Letter or A4) and margin requirements.

Japan (JP) Drawing Format

JPO requires drawing numbers in the format "【図1】" (with full-width 【】 brackets); specification text references drawings as "図1" (without brackets); reference numerals correspond to a "符号の説明" (symbol explanation) list in the specification; JP drawing format requirements are substantially identical to CN/US (black-and-white line drawings), but the full-width bracket format is a significant distinguishing feature.

Korea (KR) Drawing Format

KIPO drawing numbers use the format "도 1" (Korean "도" means "figure"); specification references write "도 1"; reference numerals use Arabic digits; format requirements are substantially similar to CN.

Jurisdiction	Drawing Number Format	Specification Reference	Numeral Format	Color Requirements
CN	图1	图1所示	Arabic (1, 2, 3)	Black-and-white by default; photographs excepted
US	FIG. 1	FIG. 1	Digit + letter (12a)	Additional request required
JP	【図1】	図1	Arabic digits	Black-and-white by default
KR	도 1	도 1	Arabic digits	Black-and-white by default
EP	Fig. 1	Fig. 1	Arabic digits	Additional request required

In CNIPA.AI, the jurisdiction configuration (JurisdictionConfig) controls the drawing reference format through the figureFormat field ("图%d" vs. "FIG. %d" vs. "【図%d】") and controls the reference numeral format through the referenceNumeralFormat field, automatically applying the correct jurisdiction-specific format in generated documents.

AI Drawing Generation: Practical Workflow

CNIPA.AI's drawing generation feature is integrated at the final stage of the patent drafting workflow. After claims and specification are complete, AI automatically reads the text content and executes the following steps:

Drawing planning: Based on the identified technical domain (automatically matched from the 7 categories), load the corresponding drawing classification specifications; list mandatory and recommended drawing types; estimate the suggested number of drawings based on the number of steps and modules in the claims (within minimum and maximum range); output the drawing list for user confirmation or adjustment.

Automated generation: Flowcharts/block diagrams/sequence diagrams — convert step descriptions in the specification to Mermaid code and render as standardized diagrams; mechanical/circuit/structural diagrams — pass corresponding descriptions as prompts to AI image generation models and generate drawings combining jurisdiction format requirements (black-and-white, line drawing); chemical structural formulas — convert SMILES strings or chemical names to standard structural formulas.

Reference numeral synchronization: Generated drawing reference numerals are automatically synchronized with descriptions in the specification, preventing "figure-text inconsistency" — the most common error in manually drawn patent drawings. Drawing captions ("Fig. N is a schematic diagram of...") are automatically generated based on drawing content.

Export integration: Generated drawings can be directly inserted into Word export documents, arranged in order ("Fig. 1, Fig. 2...") and corresponding to the reference positions in the specification body text.

In practice, for a software patent application with 10 claims, AI drawing generation reduces drawing preparation time from 4–6 hours to 15–30 minutes, while automatically ensuring complete consistency between drawing reference numerals and specification descriptions.

Common Drawing Errors: What Examiners Look For

Based on CNIPA formal examination practice, the following drawing errors are most common and most impact patentability:

Inconsistent reference numerals: Reference numeral "1" is called "base" in one part of the specification but "base plate" elsewhere; reference numeral "3" appears in a drawing but is never mentioned in the specification. These issues trigger formal examination opinions for "inconsistency between drawings and specification."

Drawings disconnected from claims: An apparatus claim mentions a "processing module" that has no corresponding block in the drawings; method claim step S3 has no corresponding node in the flowchart. Examiners look for corresponding features in the drawings while reading claims; disconnections affect comprehension.

Flowchart logical errors: A decision node (diamond) lacks both a "yes" and "no" branch; the flowchart lacks a starting point (oval "Start") and ending point ("End"); parallel steps are represented with sequential arrows. These errors cause examiners to question the completeness of the technical solution.

Overly rough drawings: Mechanical drawings lack proportional relationships with obviously distorted component dimensions; circuit symbols are non-standard; semiconductor layer diagrams cannot distinguish differences in layer thicknesses. These issues typically do not trigger formal rejection but affect quality perception and later infringement analysis.

Using AI-assisted drawing generation combined with automatic reference numeral synchronization systematically avoids the first and second types of errors; Mermaid rendering guarantees flowchart logical correctness and avoids the third type; AI image generation quality control is the improvement direction for the fourth type.

Patent Drawings and AI Generation: A Complete Guide to 7 Technical Domains, Mermaid vs. AI Imaging, and Jurisdiction Format Requirements