Case Studies
Acoustic Simulation Platform for Device Development
Key Results
- 10⁶×10⁶ matrix support
- Parallel execution on 128 GB RAM
- 30–40% memory reduction
About the project
A European manufacturer of acoustic and electronic systems tasked our team to industrialize a high-load acoustic wave and charge propagation simulation used in the development of a new smartphone. The existing solution was implemented as a MATLAB prototype and was not suitable for large-scale computation, integration into broader engineering workflows, or use by multiple teams.
The project focused on porting a computational finite-element–based model (FEM) acoustic simulation to Python and integrating it into a web-based workflow used by scientists and engineers. The system needed to support very large sparse matrices, parallel execution, and predictable performance under high memory pressure.
PerformaCode owned the work end to end. We took over the MATLAB code and mathematical models, reworked the computation pipeline, implemented parallel execution and solver strategies, and delivered a web application integrated into the customer’s engineering workflow.
8
months
5
engineers
FP
delivery model
Client challenges
The existing MATLAB-based simulation did not scale to production workloads. Large sparse matrices, high memory consumption, and solver sensitivity made execution time and resource usage unpredictable as model size increased.
This created a direct product development bottleneck. Acoustic simulation results were needed to guide device design decisions on tight hardware schedules, but long runtimes and resource-heavy execution limited the number of design variants that could be evaluated within release timelines.
The simulation was also isolated from the broader engineering workflow. Execution depended on local MATLAB environments and specialist knowledge, which constrained collaboration across teams and reduced repeatability of results.
Porting the model introduced additional technical risk. The code combined numerical methods, domain assumptions, and performance optimizations without clear separation. Any changes needed to preserve numerical correctness while materially improving memory and CPU efficiency, not just enabling parallel execution.
Tasks performed
- Reviewed and analyzed the legacy MATLAB solver and mathematical models to validate numerical methods and underlying assumptions
- Supported wave and charge propagation modeling within 3D acoustic FEM simulations
- Migrated the computational model from MATLAB to Python while preserving numerical equivalence
- Redesigned the computation pipeline for large sparse FEM workloads with explicit memory and data-flow control
- Implemented parallel computation for high-load simulations on multi-core, high-memory systems
- Optimized numerical methods for very large 3D models using sparse matrix techniques
- Reduced memory and CPU consumption through low-level performance refinements
- Evaluated and selected linear solvers using NumPy, SciPy, and sparse libraries based on stability and performance
- Implemented task orchestration and queuing for long-running simulations
- Exposed simulation control via REST APIs for integration into shared engineering workflows
- Designed and implemented a lightweight web interface for scientists and engineers
- Verified numerical stability and performance against reference models and large-scale workloads
- Prepared production-ready release artifacts for use within the customer’s engineering environment
Project results
10⁶×10⁶ matrix support
Enabled sparse 3D FEM simulation models with matrices up to 1,000,000 × 1,000,000 by redesigning data structures and memory flow.
30–40% memory reduction
Reduced peak memory usage by approximately 30–40% through sparse representations, solver tuning, and low-level allocation control.
128 GB parallel runs
Stabilized parallel execution of high-load simulations on systems with 128 GB RAM using controlled multi-core processing.
Stable runtime behavior
Achieved consistent runtime and resource usage by isolating numerical logic from execution and solver orchestration.
End-to-end execution enabled
Enabled full simulation runs through a web application with parallel task processing, replacing fragmented scripts and manual execution steps.
Value we bring
Industrializing scientific models for product use
PerformaCode has experience taking research-grade mathematical models and turning them into systems that can run reliably inside product development workflows. This includes making assumptions explicit, bounding resource usage, and aligning numerical fidelity with real delivery timelines. The result is simulation software that supports product decisions rather than remaining a specialist-only research artifact.
Optimizing hardware-level performance for numerical workloads
Our teams combine scientific computing background with low-level performance engineering. This allows us to identify where memory layout, data movement, and execution strategy matter more than algorithm changes, and to optimize accordingly. We focus on practical gains that make large-scale models runnable on available hardware rather than theoretical peak performance.
Porting legacy scientific code without losing result credibility
We know how to migrate complex scientific codebases across languages and runtimes while preserving trust in the results. This includes validating numerical equivalence, controlling solver behavior, and managing changes incrementally so that performance improvements do not come at the cost of correctness. This approach reduces adoption risk when legacy models are modernized.
Technologies
- MATLAB
- Python
- C/C++
- NumPy
- SciPy
- RabbitMQ
- REST APIs
- Linux
Other Case Studies
Edge AI Drone for Orchard Monitoring
Development of a drone-based video analytics system, delivering re...
