1.3: SYS.3 System Architectural Design


What You'll Learn

Here's what you'll take away from this section:

  • Develop system architecture from requirements
  • Allocate requirements to system elements
  • Define interfaces between elements
  • Apply AI for architecture pattern matching
  • Integrate Model-Based Systems Engineering (MBSE) tools with AI assistance
  • Design safety architectures with ASIL decomposition and freedom from interference
  • Verify architectural consistency and completeness using automated analysis
  • Establish robust requirements-to-architecture traceability with AI support

Process Definition

Purpose

ASPICE PAM v4.0 Official Purpose:

The purpose is to establish a system architectural design, identify which system requirements are allocated to which elements of the system, and to evaluate the system architectural design against defined criteria.

Key Requirements (from purpose statement):

  • Architectural design: The system must be decomposed into elements with defined relationships (BP1)
  • Static aspects: System elements, their interfaces, and structural relationships (BP1)
  • Dynamic aspects: Behavior, interaction between elements, system modes, timing (BP2)
  • Allocation: Each system requirement must be allocated to one or more system elements (BP1)
  • Evaluation: The architecture must be analyzed against defined criteria such as resource budgets, timing feasibility, and functional coverage (BP3)
  • Consistency with system requirements: Bidirectional traceability required (BP4)

Cross-Reference: SYS.3 sits between SYS.2 (System Requirements Analysis) and SYS.4 (System Integration Test). The architecture produced here drives both the downstream design processes (SWE.1, SWE.2, HWE.1) and the upstream verification strategy (SYS.4 integration test specification).

ASPICE Source: PAM v4.0 Section 4.3.3

Outcomes

Outcome Description AI Support Level
O1 A system architectural design is specified including static and dynamic aspects L1 (AI suggests decomposition and patterns, human decides)
O2 The system requirements are allocated to the system elements L1-L2 (AI proposes allocation, human validates)
O3 Consistency and bidirectional traceability are established between system requirements and system architectural design L2 (AI checks links and allocation coverage, human validates)
O4 The agreed system architecture is communicated to all affected parties L0 (Human only -- ASPICE accountability)

O1 Components (for pedagogical clarity -- these are NOT separate outcomes):

  • Static aspects (BP1):
    • System elements identified (SW, HW, mechanical, external)
    • Interfaces between elements defined (HSI, communication buses, physical connectors)
    • Relationships and dependencies documented
    • Requirements allocated to elements
  • Dynamic aspects (BP2):
    • System behavior in operational modes (Init, Normal, Degraded, Error, Sleep)
    • Interaction sequences between elements (timing, data flow, control flow)
    • Resource sharing and arbitration mechanisms

O2 Allocation Principles:

  • Every system requirement must be allocated to at least one system element
  • A single requirement may be allocated to multiple elements (e.g., SW + HW)
  • Allocation rationale must be documented for each assignment
  • Derived requirements at the element level must trace back to the originating system requirement

Base Practices with AI Integration

IMPORTANT: SYS.3 has exactly 5 base practices. The table below uses descriptions aligned with PAM v4.0.

BP Official Base Practice Description AI Level AI Application HITL Required
BP1 Specify static aspects of the system architecture. Identify the system elements and their interfaces, and document the static aspects of the system architecture with respect to functional and non-functional system requirements. Allocate system requirements to the system elements. L1 AI suggests element decomposition based on requirement clustering, proposes allocation using pattern matching from similar projects. Human validates all architectural decisions. YES -- Human validates decomposition and allocation
BP2 Specify dynamic aspects of the system architecture. Specify and document the dynamic aspects of the system architecture with respect to the functional and non-functional system requirements, including system modes, interaction between system elements, and timing aspects. L1 AI suggests behavioral models and timing budgets, generates interaction sequence drafts from requirements. Human validates dynamic behavior and timing. YES -- Human validates modes and timing
BP3 Analyze system architecture. Analyze the system architecture against defined criteria such as functionality coverage, resource consumption, and timing feasibility. Document a rationale for the system architectural design decisions. L1-L2 AI performs technical analysis (resource budgets, timing feasibility, coverage metrics), identifies potential bottlenecks and conflicts. Human validates findings and provides design rationale. YES -- Human provides design rationale
BP4 Ensure consistency and establish bidirectional traceability. Ensure consistency and establish bidirectional traceability between the system architectural design and the system requirements. L2 AI checks trace links and allocation completeness, detects orphan requirements and unlinked elements. Human validates critical allocation decisions. YES -- Human validates allocation completeness
BP5 Communicate agreed system architecture. Communicate the agreed system architecture to all affected parties. L0 Human only -- ASPICE requires human accountability for agreements and communication of architectural decisions. YES -- Human communication required

PAM Notes on Static Aspects (BP1):

  • Note 1: System elements can include hardware elements, software elements, mechanical elements, and external elements (sensors, actuators, communication buses)
  • Note 2: The allocation should consider both functional and non-functional requirements (performance, safety, security)

PAM Notes on Dynamic Aspects (BP2):

  • Note 3: Examples of system modes include: Initialization, Normal Operation, Degraded Mode, Error Handling, Sleep/Low Power
  • Note 4: Dynamic aspects should address timing budgets, data flow rates, and event-driven interactions between system elements

PAM Notes on Analysis (BP3):

  • Note 5: Examples of criteria to analyze: functionality coverage, resource consumption (CPU, memory, bandwidth), timing feasibility, safety requirements allocation, testability
  • Note 6: Design rationale can include: proven-in-use arguments, reuse of platform components, technology trade-offs, cost optimization, safety decomposition justification

ASPICE Source: PAM v4.0 Section 4.3.3


AI in Architectural Design

AI-Assisted Architecture Generation

AI can support system architects by analyzing the full set of system requirements and proposing candidate architectures. This goes beyond simple pattern matching -- it involves requirement clustering, constraint analysis, and technology mapping.

AI Architecture Generation Pipeline:

Step 1: Requirement Analysis
    AI Input:  Full set of system requirements (SYS-REQ-xxx)
    AI Action: Cluster requirements by functional domain, quality attribute,
               and allocation target (SW, HW, mechanical, external)
    AI Output: Requirement clusters with suggested element boundaries

Step 2: Architecture Pattern Selection
    AI Input:  Requirement clusters + project constraints (ASIL level, cost,
               platform reuse, existing IP)
    AI Action: Match against known architecture patterns from reference database
    AI Output: Ranked list of candidate architecture patterns with rationale

Step 3: Element Definition
    AI Input:  Selected pattern + requirement allocation constraints
    AI Action: Propose concrete system elements with responsibilities,
               interfaces, and requirement allocation
    AI Output: Draft element table, interface list, allocation matrix

Step 4: Human Review and Decision
    Architect: Evaluates AI proposals against domain knowledge, project
               constraints, organizational standards, and safety requirements
    Decision:  Accept / Modify / Reject each AI suggestion

AI-Assisted Architecture Evaluation

Evaluation Criterion AI Capability Confidence Human Role
Functionality coverage Checks every requirement is allocated to at least one element High Validate edge cases
Resource estimation Estimates CPU load, memory usage, bus bandwidth from similar projects Medium Validate against actual hardware specs
Timing feasibility Analyzes timing chains from stimulus to response across elements Medium Validate against real-time constraints
Safety allocation Checks ASIL decomposition rules, flags potential freedom-from-interference issues Medium Validate safety arguments
Interface completeness Detects missing interfaces between elements that share data or control High Validate physical feasibility
Testability Assesses whether each element can be independently verified Medium Validate test infrastructure availability

Key Principle: AI excels at exhaustive checking (coverage, completeness, consistency) but cannot replace human judgment on trade-offs, stakeholder priorities, or novel design decisions. The architect retains full decision authority.


Architecture Patterns

Common Patterns for Embedded Systems

The following table summarizes architecture patterns frequently applied in automotive and industrial embedded systems, along with AI integration points at each level.

Pattern Description When to Use AI Integration Point
Layered Architecture Strict separation into Application, Service, Driver, and Hardware Abstraction layers General-purpose ECU software (AUTOSAR-compliant systems) AI suggests layer assignment for each element based on coupling analysis
Federated Architecture Independent ECUs connected via communication bus, each owning a functional domain Vehicle-level systems with clear domain boundaries (powertrain, body, chassis) AI analyzes requirement dependencies to propose ECU boundaries
Centralized Architecture Single high-performance compute node executing multiple functions Domain controllers, zonal architectures, advanced driver assistance AI evaluates resource budgets and partitioning feasibility
Redundant Architecture Duplicated elements with voting or comparison for fault tolerance Safety-critical functions (steering, braking) requiring ASIL C/D AI checks redundancy coverage against safety requirements
Sensor-Controller-Actuator Classic control loop pattern: sense, decide, act Real-time control functions (motor control, climate, lighting) AI maps requirements to sensor/controller/actuator elements
Client-Server Service-oriented communication between provider and consumer elements Diagnostic services, calibration interfaces, OTA update AI identifies service interfaces from requirement analysis

Pattern Selection Criteria

Criterion Weight AI Analysis Approach
Safety integrity level (ASIL) High Pattern must support required ASIL decomposition and FFI
Real-time requirements High Pattern must support timing budgets and deterministic behavior
Reuse and platform constraints Medium Pattern should align with existing platform IP and AUTOSAR baseline
Development cost Medium Pattern complexity impacts effort estimation
Communication bandwidth Medium Pattern determines bus load and data flow topology
Scalability Low-Medium Pattern should accommodate future feature growth

Practical Guidance: For most automotive body control functions (BCM, lighting, comfort), the Layered Architecture pattern combined with Sensor-Controller-Actuator is the default starting point. AI should propose this as the baseline and flag deviations only when requirements demand it (e.g., safety-critical functions requiring redundancy).


System Architecture Example

Block Diagram

The block diagram below shows the BCM system architecture, identifying hardware, software, and external elements along with their interconnections.

BCM System Architecture

Element Definitions

Element Type Function
Microcontroller Hardware Execute BCM software
BCM Application SW Software Implement control logic
Power Management Hardware Voltage regulation, protection
Driver Circuits Hardware Interface to actuators
CAN Interface Hardware Vehicle network communication
Door Lock Motors External Physical door locking
Sensors External Position and status feedback

Requirements Allocation

Allocation Matrix

The following diagram illustrates the ID naming convention used for requirement allocation, ensuring consistent traceability across hardware, software, and external elements.

ID Convention

Allocation Rationale

Requirement Allocation Rationale
Lock timing (SYS-BCM-010) SW + HW + External Timing budget distributed
CAN communication SW + HW Protocol in SW, physical in HW
Power management HW only Analog circuit function
User input handling SW only Logic decision in software

MBSE Integration

Using AI with Model-Based Architecture Tools

Model-Based Systems Engineering (MBSE) is the standard practice for system architectural design in automotive and aerospace domains. Tools such as Eclipse Capella, IBM Rhapsody, and Cameo Systems Modeler provide formal modeling environments. AI integration with these tools creates a powerful combination for SYS.3.

Cross-Reference: See Part III (AI Toolchain) for detailed tool qualification requirements when AI generates or modifies MBSE model artifacts.

AI Integration Points with MBSE Tools

MBSE Tool Model Type AI Integration Benefit
Eclipse Capella Arcadia methodology (Operational, System, Logical, Physical) AI proposes Logical Architecture decomposition from Operational Analysis; checks consistency between Capella layers Faster iteration from operational needs to logical components
IBM Rhapsody SysML (BDD, IBD, Activity, Sequence, State Machine) AI generates draft SysML Block Definition and Internal Block Diagrams from requirements; validates port/flow consistency Reduced manual modeling effort; fewer interface errors
Cameo Systems Modeler SysML + custom profiles AI assists with requirement allocation to SysML blocks; generates parametric diagrams for constraint analysis Automated allocation coverage checking
MATLAB/Simulink Dataflow models, Stateflow AI suggests model structure from requirements; generates Simulink block hierarchy drafts Rapid prototyping of dynamic behavior

MBSE-AI Workflow

Step 1: Requirements Import
    Tool:   Import system requirements (from DOORS, Polarion, Jama) into MBSE model
    AI:     Validate import completeness, flag missing attributes

Step 2: Logical Architecture Generation
    Tool:   Create logical component model (Capella Logical Architecture / SysML BDD)
    AI:     Propose component decomposition based on requirement clustering
    Human:  Review and adjust logical architecture

Step 3: Interface Definition
    Tool:   Define ports, flows, and interfaces between components (SysML IBD)
    AI:     Suggest interface data types, flow directions, and protocols from
            requirement analysis and similar project patterns
    Human:  Validate interface definitions against physical constraints

Step 4: Allocation and Traceability
    Tool:   Allocate requirements to logical/physical components
    AI:     Check allocation completeness, flag unallocated requirements
    Human:  Validate allocation decisions, document rationale

Step 5: Model Verification
    Tool:   Run model-level checks (Capella validation rules, SysML consistency)
    AI:     Augment tool checks with cross-model consistency analysis
    Human:  Review verification results, resolve findings

Practical Note: AI integration with MBSE tools is most effective when the AI has access to the tool's API or export format (e.g., Capella .melodymodeller, SysML XMI, Simulink .slx). Direct model manipulation through tool APIs provides tighter integration than working with exported documents.


Safety Architecture

ASIL Decomposition

When system requirements carry an Automotive Safety Integrity Level (ASIL) classification per ISO 26262, the system architecture must address safety decomposition. ASIL decomposition allows splitting a high-ASIL requirement across redundant or independent elements, each carrying a lower ASIL.

Cross-Reference: See ISO 26262 Part 9 (ASIL-oriented and safety-oriented analyses) for detailed decomposition rules.

Original ASIL Decomposition Option 1 Decomposition Option 2 Decomposition Option 3
ASIL D ASIL D(D) -- no decomposition ASIL C(D) + ASIL A(D) ASIL B(D) + ASIL B(D)
ASIL C ASIL C(C) -- no decomposition ASIL B(C) + ASIL A(C) ASIL A(C) + ASIL A(C) + ASIL A(C)
ASIL B ASIL B(B) -- no decomposition ASIL A(B) + ASIL A(B) --
ASIL A ASIL A -- no decomposition QM + QM (not valid) --

Important: The notation ASIL X(Y) means the element is developed to ASIL X, but the decomposition was derived from an original ASIL Y requirement. The original ASIL in parentheses must be maintained in traceability records.

Freedom from Interference (FFI)

Freedom from interference is a fundamental safety architecture principle. It requires that elements of different ASIL levels (or QM elements coexisting with ASIL elements) cannot negatively influence each other's correct operation.

FFI Mechanism Description Applicable To AI Assistance
Memory partitioning Hardware or MPU-enforced memory separation between elements SW elements of different ASIL AI checks partition configuration completeness against element allocation
Temporal partitioning Time-triggered scheduling ensuring no timing interference SW tasks of different ASIL on shared CPU AI analyzes WCET budgets and scheduling feasibility
Communication protection CRC, sequence counters, alive counters on inter-element messages Communication between elements of different ASIL AI validates that all safety-relevant interfaces have appropriate E2E protection
Hardware separation Physical isolation of hardware elements (separate power supply, separate PCB areas) HW elements of different ASIL AI checks that hardware block diagram reflects required physical separation
Monitoring and voting Independent monitor detects failures in primary element Redundant architectures for ASIL C/D AI validates that monitor elements have independent failure modes

AI Assistance for Safety Architecture

AI Safety Architecture Analysis:
─────────────────────────────────
Input:  System requirements with ASIL classifications
        Proposed system architecture (element decomposition)

Analysis Performed:
1. ASIL Inheritance Check
   - Every element inherits the highest ASIL of its allocated requirements
   - Flag elements with mixed-ASIL allocation (potential FFI concern)

2. Decomposition Validation
   - Verify decomposition follows ISO 26262 Part 9 rules
   - Check that dependent failures are addressed
   - Flag decomposition without sufficient independence argument

3. FFI Gap Analysis
   - Identify element pairs that share resources (memory, CPU, bus)
   - Check that FFI mechanisms are specified for each shared resource
   - Flag missing FFI mechanisms

4. Safety Requirement Allocation Coverage
   - Verify all safety requirements are allocated to elements
   - Check that safety mechanisms (monitoring, watchdog, diagnostic)
     are allocated to specific elements

Human Review Required:
- All ASIL decomposition decisions
- FFI argument sufficiency
- Safety mechanism adequacy
- Dependent failure analysis conclusions

Key Principle: AI can identify potential safety architecture gaps with high confidence through exhaustive rule-based checking. However, the safety case argumentation -- demonstrating that the architecture is sufficiently safe -- requires human engineering judgment and remains a human responsibility.


Interface Design

AI-Assisted Interface Definition Between System Elements

Interface design is one of the most error-prone activities in system architecture. Missing interfaces, inconsistent data types, and undocumented timing constraints account for a significant portion of integration defects. AI can substantially reduce these errors.

Interface Specification

Door Lock Driver Interface

Note: Register addresses and bit assignments are illustrative; actual values are project-specific.

Register Address Description
DOOR_LOCK_CTRL Base + 0x100 Door lock control register

Bit Assignment:

Bit Name Description
7-4 Reserved Read as 0
3 RR_LOCK Rear Right (1=lock, 0=unlock)
2 RL_LOCK Rear Left (1=lock, 0=unlock)
1 FR_LOCK Front Right (1=lock, 0=unlock)
0 FL_LOCK Front Left (1=lock, 0=unlock)

Timing:

Operation Maximum Time
Write to register < 1us
Command to actuator < 5ms
Actuator feedback < 200ms

Error Handling:

Condition Response
Actuator stuck Timeout after 1s
Overcurrent Hardware protection, status bit set

Hardware-Software Interface (HSI)

Abbreviation: HSI = Hardware-Software Interface. See also Part I glossary.

The HSI is the most critical interface in embedded system architecture. It defines how software accesses hardware resources and must be specified with sufficient detail to enable independent HW and SW development.

HSI Category Content AI Generation Capability
Register map Address, bit fields, access mode (R/W/RW), reset values L2 -- AI generates register tables from datasheet analysis
Interrupt map Interrupt sources, priorities, vectors, latency requirements L1-L2 -- AI suggests interrupt allocation from timing requirements
Memory map Address ranges, memory types (Flash, RAM, EEPROM), access permissions L1 -- AI proposes memory partitioning from element allocation
Timing constraints Setup/hold times, clock domains, synchronization requirements L1 -- AI extracts timing constraints from datasheets and requirements
Electrical characteristics Voltage levels, current limits, pin assignments L0-L1 -- Primarily human, AI assists with datasheet extraction

External Interface

Interface Type Characteristics
CAN Bus Communication 500 kbps, ISO 11898
Door Switches Digital Input Active low, 10ms debounce
Lock Motors PWM Output 12V, 5A max, 20 kHz
Position Sensors Analog Input 0-5V, 10-bit ADC

AI-Assisted Interface Consistency Checking

Check Type AI Action Detection Capability
Data type mismatch Compare data types on both sides of each interface Detects when sender uses uint16 but receiver expects uint8
Unit mismatch Verify physical units match across interface boundaries Detects when sender provides value in mV but receiver expects V
Range mismatch Check that value ranges are compatible across interfaces Detects when sender range [0..4095] exceeds receiver range [0..255]
Timing mismatch Compare sender update rate with receiver expected rate Detects when sender updates at 100ms but receiver expects 10ms
Missing interface Analyze data flow requirements to find unconnected elements Detects when two elements share a requirement but have no defined interface
Bidirectional completeness Verify that request-response pairs are complete Detects when a command interface exists but no feedback interface is defined

Practical Guidance: Run AI interface consistency checks after every architectural change. Interface errors found during architecture design cost 10-100x less to fix than interface errors found during system integration testing (SYS.4).


Architecture Verification

Automated Consistency Checking

Architecture verification ensures that the system architectural design is internally consistent, complete with respect to system requirements, and feasible with respect to technical constraints. AI-assisted verification addresses three dimensions.

Verification Dimension What Is Checked AI Approach Confidence
Internal consistency No contradictions within the architecture (e.g., element A sends data to B, but B has no receiving port) Rule-based analysis of element-interface graph High
Completeness Every system requirement is allocated; every element has at least one requirement; every interface is fully specified Coverage analysis against requirement set High
Feasibility Timing budgets are achievable; resource budgets (CPU, memory, bus bandwidth) are within limits Estimation from similar projects and analytical models Medium

Completeness Analysis

Completeness Check Pass Criterion AI Detection Method
Requirement allocation coverage 100% of system requirements allocated to at least one element Set comparison: requirements vs. allocation matrix
Element justification Every element has at least one allocated requirement Reverse check: elements without requirements flagged
Interface specification Every interface has data type, direction, timing, and error handling defined Attribute completeness check on interface table
Dynamic behavior coverage Every system mode is described with entry/exit conditions and element behavior Mode table completeness check
Safety requirement allocation Every ASIL-rated requirement is allocated with FFI consideration Safety attribute check on allocation matrix

Architecture Analysis Report

The architecture analysis report (BP3) documents the evaluation results and design rationale. AI assists in generating this report.

AI-Generated Architecture Analysis Report (Draft):
───────────────────────────────────────────────────
1. Functionality Coverage
   - 47/47 system requirements allocated                    [PASS]
   - 3 derived requirements identified at element level     [INFO]
   - 0 orphan requirements                                  [PASS]

2. Resource Estimation
   - CPU load estimate: 62% (budget: 80%)                   [PASS]
   - RAM usage estimate: 38KB (budget: 64KB)                [PASS]
   - Flash usage estimate: 180KB (budget: 256KB)            [PASS]
   - CAN bus load estimate: 45% (budget: 70%)               [PASS]

3. Timing Analysis
   - Door lock end-to-end: 145ms (budget: 200ms)            [PASS]
   - CAN response time: 12ms (budget: 50ms)                 [PASS]
   - Sleep-to-wake transition: 85ms (budget: 100ms)         [WARN]
     Margin: 15ms -- recommend review of wake-up sequence

4. Safety Analysis
   - ASIL B requirements allocated to ASIL B elements       [PASS]
   - FFI mechanisms specified for all mixed-ASIL boundaries  [PASS]
   - 1 interface between QM and ASIL B missing E2E check    [FAIL]
     Action: Add CRC protection to interface IF-BCM-017

5. Design Rationale Summary
   - Layered architecture selected: proven-in-use from BCM platform v2.3
   - CAN protocol stack: reuse from AUTOSAR BSW
   - Motor driver: dual-channel for diagnostic capability

Human Action: Review all findings, resolve FAIL items, approve report

Traceability

Requirements-to-Architecture Traceability with AI

Traceability between system requirements and the system architecture is an explicit ASPICE outcome (O3) and base practice (BP4). AI provides substantial value by automating the creation, maintenance, and verification of trace links.

Traceability Structure

Trace Link Type From To Purpose
Allocation System requirement (SYS-REQ-xxx) System element (SE-xxx) Shows which element implements each requirement
Derivation System requirement (SYS-REQ-xxx) Derived requirement (DRV-REQ-xxx) Captures requirements created during architecture decomposition
Interface trace System requirement (SYS-REQ-xxx) Interface specification (IF-xxx) Links communication/timing requirements to specific interfaces
Verification trace System element (SE-xxx) Integration test case (SYS-IT-xxx) Links architecture to verification (forward to SYS.4)

AI-Assisted Traceability Operations

Operation Manual Effort AI-Assisted Effort AI Approach
Initial trace creation 2-4 hours per 50 requirements 20-30 minutes (AI proposes, human validates) NLP matching of requirement text to element descriptions
Trace maintenance after change 30-60 minutes per change 5-10 minutes (AI identifies impacted traces) Impact analysis on requirement-element graph
Coverage gap detection 1-2 hours (manual review) Seconds (automated check) Set comparison of requirements vs. allocation matrix
Consistency validation 1-2 hours (manual cross-check) Seconds (automated check) Rule-based consistency analysis on trace links
Trace report generation 1-2 hours (manual formatting) Minutes (AI generates report) Template-based report from trace data

AI Traceability Validation Results (Example)

Traceability Check Results:
───────────────────────────
Total system requirements:     47
Allocated requirements:        47/47  (100%)                 [PASS]
Requirements with rationale:   45/47  (95.7%)                [WARN]
  Missing rationale: SYS-BCM-042, SYS-BCM-046

Bidirectional check:
  Upward:  47/47 requirements trace to stakeholder needs     [PASS]
  Downward: 47/47 requirements allocated to elements         [PASS]

Element coverage:
  Total elements: 7
  Elements with requirements: 7/7                            [PASS]
  Elements with >10 requirements: 2 (BCM Application SW, Microcontroller)
    Recommendation: Review for potential decomposition

Derived requirements:
  Total derived: 3
  Derived with upward trace: 3/3                             [PASS]

Interface traces:
  Interfaces with requirement links: 12/14                   [WARN]
  Missing traces: IF-BCM-013 (internal bus), IF-BCM-014 (debug port)
  Action: Verify whether these interfaces relate to any system requirement

Human Action: Review warnings, add missing rationale, validate interface traces

AI Integration for Architecture

L1: Pattern Suggestions

AI Input: System requirements for door lock control

AI Analysis:
─────────────
Based on similar automotive BCM architectures, I suggest:

1. Architecture Pattern: Layered with Hardware Abstraction
   - Application Layer (door lock logic)
   - Service Layer (timing, diagnostics)
   - Driver Layer (hardware abstraction)
   - Hardware Layer (registers, peripherals)

2. Similar Projects Reference (hypothetical examples):
   - Project BCM-A: 200ms requirement achieved with ARM Cortex-M4
   - Project BCM-B: Used dual-channel motor driver for redundancy

3. Potential Issues:
   - EMC: Motor noise may affect CAN communication
   - Timing: Simultaneous 4-door lock may exceed current limit

Human Action: Evaluate suggestions, adapt to project context

L2: Documentation Generation

AI can generate draft architecture documentation including:
- Block diagrams (Mermaid/PlantUML syntax)
- Interface tables
- Allocation matrices
- Traceability links

Human reviews and approves all generated content

HITL Protocol for Architectural Decisions

Decision Authority Matrix

Every architectural decision made during SYS.3 must follow the Human-in-the-Loop (HITL) protocol. The following table defines who decides, who assists, and what evidence is required.

Decision Category AI Role Human Role Decision Authority Evidence Required
System element decomposition Proposes candidate decompositions based on requirement clustering Evaluates trade-offs, selects decomposition Architect (human) Architecture description with rationale (04-06)
Requirement allocation Proposes allocation based on element capabilities and pattern matching Validates allocation, resolves conflicts Architect (human) Allocation matrix with rationale
Interface definition Generates draft interface specifications from requirements and datasheets Validates data types, timing, protocols Architect + domain expert (human) Interface specification (04-03)
ASIL decomposition Checks decomposition rules, flags violations Decides decomposition strategy, argues sufficiency Safety engineer (human) Safety analysis report
Technology selection Compares options against requirements and constraints Decides technology based on cost, availability, organizational strategy Architect + project manager (human) Design rationale document
Architecture approval Generates review package with analysis results Reviews, approves, communicates to stakeholders Architecture review board (human) Review record, sign-off

HITL Escalation Protocol

Situation AI Action Human Action
AI confidence below 60% on allocation suggestion Flags low confidence, presents alternatives Must manually evaluate and decide
Conflicting requirements detected Identifies conflict, proposes resolution options Must resolve with stakeholders
Safety requirement affected by change Flags impact on safety case, stops automated processing Safety engineer must re-evaluate
Novel requirement without pattern match Reports no matching pattern, provides closest analogies Must design from first principles
Resource budget exceeded (>80% utilization) Flags risk, suggests optimization options Must decide on architecture modification or budget increase

Non-Negotiable Rule: No architectural decision that affects safety (ASIL allocation, FFI mechanism, redundancy strategy) shall be made by AI alone. The safety engineer must explicitly approve every safety-related architectural decision, and this approval must be recorded in the project's safety case documentation.


Work Products

WP ID Work Product Content AI Role HITL
04-06 System architecture description Elements, interfaces, behavior, system modes, resource budgets AI generates draft descriptions from requirements and patterns; human reviews, modifies, and approves Architect approves
17-08 Updated SRS Allocation annotations, derived requirements added during architecture AI identifies derived requirements and proposes allocation annotations; human validates Architect validates
17-11 Traceability record Bidirectional links between system requirements and system elements AI generates trace links and checks coverage; human validates critical links Architect validates
04-03 Interface specification HSI, external interfaces, communication protocols, timing AI generates draft interface tables from requirements and datasheets; human validates Architect + HW engineer validate
15-51 Analysis results Architecture evaluation against criteria (functionality, timing, resources, safety) AI performs automated analysis and generates report; human validates findings and provides rationale Architect approves
13-51 Consistency evidence Proof of bidirectional traceability, allocation completeness AI generates traceability matrices and coverage reports; human validates completeness Architect validates
13-52 Communication evidence Stakeholder approvals, review records, agreement documentation Human only -- meeting minutes, sign-offs, distribution records Architect + stakeholders

ASPICE Compliance Note: Work products 04-06 (System Architecture Description) and 04-03 (Interface Specification) are the primary outputs of SYS.3. These must include BOTH static and dynamic aspects (O1) and demonstrate complete requirement allocation (O2). Work products 13-51 and 17-11 provide the traceability evidence required by O3.


Implementation Checklist

Use this checklist to verify that all SYS.3 activities have been completed for ASPICE compliance. Items marked with "(AI)" indicate where AI can assist; items marked with "(Human)" require human judgment and decision.

BP1: Static Aspects

Step Activity AI/Human Status
1.1 Identify all system elements (SW, HW, mechanical, external) AI proposes, Human decides
1.2 Define element responsibilities and boundaries Human decides, AI documents
1.3 Allocate each system requirement to at least one element AI proposes, Human validates
1.4 Document allocation rationale for each assignment Human writes, AI assists with templates
1.5 Define all interfaces between elements AI identifies from data flow, Human validates
1.6 Specify interface data types, directions, and protocols AI generates draft, Human validates

BP2: Dynamic Aspects

Step Activity AI/Human Status
2.1 Define system operational modes (Init, Normal, Degraded, Error, Sleep) Human defines, AI checks completeness
2.2 Specify mode transitions with entry/exit conditions Human defines, AI validates consistency
2.3 Document interaction sequences between elements AI generates drafts, Human validates
2.4 Define timing budgets for end-to-end scenarios Human defines budgets, AI checks feasibility
2.5 Address resource sharing and arbitration mechanisms Human designs, AI checks for conflicts

BP3: Architecture Analysis

Step Activity AI/Human Status
3.1 Verify functionality coverage (all requirements allocated) AI automated check
3.2 Estimate resource consumption (CPU, RAM, Flash, bus load) AI estimates, Human validates
3.3 Analyze timing feasibility for critical paths AI analyzes, Human validates
3.4 Evaluate safety architecture (ASIL decomposition, FFI) AI checks rules, Human validates arguments
3.5 Document design rationale for architectural decisions Human writes, AI assists with structuring
3.6 Generate architecture analysis report AI generates draft, Human reviews and approves

BP4: Traceability

Step Activity AI/Human Status
4.1 Create bidirectional trace links (requirements to elements) AI generates, Human validates
4.2 Verify 100% allocation coverage AI automated check
4.3 Document derived requirements with upward traces AI identifies, Human validates
4.4 Verify trace consistency (no broken links) AI automated check
4.5 Generate traceability matrix report AI generates report

BP5: Communication

Step Activity AI/Human Status
5.1 Prepare architecture review package AI generates package, Human reviews
5.2 Conduct architecture review with stakeholders Human only
5.3 Document review decisions and action items Human documents, AI assists with formatting
5.4 Obtain formal approval and sign-off Human only
5.5 Distribute approved architecture to all affected parties Human distributes, AI generates distribution record

Summary

SYS.3 System Architectural Design:

  • AI Level: L1 (AI assists, human decides)
  • AI Value: Pattern matching, documentation generation
  • Human Essential: Architecture decisions, trade-offs
  • Key Outputs: Architecture description, allocation, interfaces
  • Critical Element: Hardware-software interface specification

AI Integration by Base Practice:

  • BP1 (Static aspects): L1 -- AI proposes element decomposition and allocation, human validates all architectural decisions
  • BP2 (Dynamic aspects): L1 -- AI suggests behavioral models and timing budgets, human validates modes and interactions
  • BP3 (Analysis): L1-L2 -- AI performs automated feasibility and coverage analysis, human validates findings and documents rationale
  • BP4 (Traceability): L2 -- AI generates and checks trace links, human validates allocation completeness
  • BP5 (Communicate): L0 -- Human only (ASPICE accountability for agreements and communication)

Human-in-the-Loop (HITL) Requirements:

  • ALL base practices require human review and validation
  • BP5 (agreement/communication) is human-only (L0)
  • Safety-related architectural decisions (ASIL decomposition, FFI, redundancy) require explicit safety engineer approval
  • AI assists with analysis, checking, and documentation -- humans make all final architectural decisions