Technology module
On-chip data fabric
As compute units multiply, system efficiency depends on data arriving at the right place at the right time. The VXI on-chip data fabric coordinates data paths, traffic, and module communication around each workload so the whole architecture behaves as one system.
Architecture · Dataflow · Integration
Why it matters
Coordinate data paths, traffic governance, and module communication around each workload.
Workload-aware data paths
Transfer paths follow the data needs of each model stage, reducing congestion and stalls.
Traffic governance and QoS
Different data streams get their own priorities so critical inference tasks stay responsive.
Scalable modular integration
Consistent interfaces link compute, memory, and peripheral modules, supporting a family of platforms.
How it works
Traffic analysis
Characterize dataflow patterns and bandwidth hot spots during model execution.
Path planning
Assign the best routes for weights, activations, and control messages.
Dynamic scheduling
Coordinate streams at runtime by task priority to avoid resource contention.
System co-optimization
Tune end-to-end efficiency together with the memory hierarchy and runtime.
What we align on during evaluation
Bring this information to your evaluation request so the VXI team can propose a concrete platform and adoption plan faster.
- Number of concurrent tasks and their dataflow patterns
- Real-time and quality-of-service needs of critical tasks
- Connections to peripherals, sensors, or companion accelerators
- Future expansion and platform family plans
Frequently asked questions
- Why does on-chip interconnect matter?
- In AI silicon, moving data often costs more time and energy than computing on it. A well-designed fabric lets modules work together, cutting stalls and raising overall utilization.
- How does the platform stay stable with several AI tasks at once?
- VXI coordinates data streams with traffic governance and priority mechanisms, keeping critical tasks responsive while background work continues.
- How does the fabric support different product platforms?
- Modular fabric interfaces integrate compute, memory, and peripherals consistently, helping one architecture extend across multiple product platforms.
Related technology modules
- SRAM-CIM compute path — Weight residency and near-memory compute that cut data movement
- Memory hierarchy — Co-placed weights, activations, and cache
- System co-engineering — An integration path connecting model, software, host, and memory