Optimizing Complex Mathematics for Hardware Execution

Full Metal Analytics is a deep-tech startup specializing in the design and validation of hardware-optimized numerical kernels. We bridge the gap between complex mathematical theory and silicon-ready implementations.


Founded by PhDs in Applied Mathematics, we are currently validating high-performance algorithms through resource-constrained software prototyping to ensure optimal efficiency for future hardware integration.

Logo

Our Development Path

01

Algorithmic Discovery

Deriving rigorous mathematical models for complex dynamics.

02

Resource-Constrained Validation

Prototyping via web technologies and low-level software to profile performance.

03

Hardware Integration

Transitioning validated IP into specialized silicon and EDA flows.

Numerical IP Design

Creating stable, high-precision kernels designed for minimal silicon footprint and maximum throughput.

Hardware-Software Co-Design

Bridging the gap between mathematical complexity and low-level hardware constraints.

Performance Profiling

Rigorous benchmarking of algorithmic efficiency in resource-constrained environments.

Targeted Computational Domains

Real-Time Simulation | Semiconductor EDA | High-Performance Computing | Quantitative Finance

Validation Prototypes

Current research focus: testing mathematical stability and computational efficiency in resource-constrained environments.

Risk Management

Quant-Kernel Validation

Prototyping pricing algorithms for low-latency environments.

Automatic Differentiation

Gradient Engine Prototyping

Testing automatic differentiation for hardware acceleration.

Development Progress

AAD Kernel Validation

25 Aug 2025

Benchmarking alpha-release of our adjoint engine.

Web-Based Prototyping

29 Jul 2025

Successful implementation of WASM-based PDE solvers for rapid algorithmic prototyping.

Products

Screenshot Screenshot Screenshot

Risk Management Live

A simple application for managing trading positions, calculating risk sensitivities and pricing financial instruments.

  • Optimized numerical algorithms for fast and efficient calculations.
  • Integrate your own mathematical models and financial instruments.
  • Designed to collaborate seamlessly with AWS or on-premises cloud.

Automatic Differentiation Live

An educational application for generating automatic differentiation code.

  • Generate tangent and adjoint code using Python-like syntax.
  • Runs entirely client-side in the browser.
Screenshot Screenshot

Risk Management Live

Screenshot

A simple application for managing trading positions, calculating risk sensitivities and pricing financial instruments.

  • All-in-one browser application engineered for performance.
  • Optimized numerical algorithms for fast calculations.
  • Includes support for Black-Scholes, Dupire and Heston.
Screenshot Screenshot

About

Logo

Full Metal Analytics is a partnership of PhDs in mathematics specializing in high-performance numerical methods.

We focus on the practical implementation of mathematical theory, specifically designing numerical kernels that are optimized for hardware execution. Our goal is to bridge the gap between complex mathematical models and the constraints of silicon-ready software and hardware.


Numerical Implementation

Our expertise includes the development of PDE solvers using adaptive mesh refinement and finite element methods. We focus on the numerical linear algebra required to make these solvers efficient, selecting algorithms and preconditioning strategies that balance computational accuracy with speed.


Hardware-Software Co-Design

By combining mathematical expertise with a background in low-level systems and electronics, we approach problems through a hardware-aware lens. We work to ensure that complex simulations can run in real-time on specialized hardware, providing both consulting and validated IP for demanding computational environments.


Contact us to discuss a technical collaboration →

Technical Inquiries

Direct contact for consulting, licensing, or research collaboration.

By submitting, you consent to our processing of your data for the purpose of responding to your inquiry. Privacy Policy.

Logo

Validation Prototypes

Real-time implementations demonstrating mathematical stability and computational efficiency within resource-constrained browser environments.

Powered by modern web technologies

Moving Mesh

2D PDE Solving

1D PDE Solving

Automatic Adjoint Differentiation

Responsive Mesh Dynamics

Mesh adaptation is a vital mathematical tool for solving complex, high-gradient problems by concentrating computational resources where they are needed most. This demo was developed by our team of numerical analysts to showcase our expertise in advanced PDE solving.

Presets

Mathematical Settings

Visual Styles

Grid Colour
Color Mode
Chemtrail
Chessboard
Advanced settings
Boundary Conditions
Jacobi x0

Reaction-Diffusion: Adaptive Dynamics

Unlike traditional h or p-refinement, this demo utilizes an evolving moving mesh to concentrate resolution on high-gradient features. By optimizing the ratio of accuracy to degrees of freedom, we can achieve high-fidelity results with significantly fewer mesh points, carefully balancing mesh-update overhead against computational savings.

The Gray-Scott reaction-diffusion pattern generating model

Initial Conditions

Reaction Parameters

Mesh Settings

Moving mesh apation
Show grid

High-Speed 1D ADE Solver

. Simulation failed.




This solver implements an Alternating-Direction-Explicit (ADE) scheme to achieve high-speed simulation of advection-diffusion-reaction dynamics. The lack of a global dependency (which occurs in implicit methods where every point depends on every other point through the matrix solve) allows for massive scaling across thousands of processors, enabling rapid, real-time parameter exploration.

Domain: t > 0, xleft ≤ x ≤ xright

∂u/∂t = A ∂²u/∂²x + B ∂u/∂x + Cu + D

Initial: u(t=0, x) = u0(x)

Left BC: α0u + β0u' + γ0u'' = ε0

Right BC: αNu + βNu' + γNu'' = εN

Automatic Adjoint Differentiation (AAD)

Automatic Adjoint Differentiation (AAD) provides a mathematically exact way to compute sensitivities (gradients) without the prohibitive cost of finite difference methods. By propagating derivatives backward through the computational graph, AAD enables high-dimensional optimization and parameter estimation with a computational cost that is nearly independent of the number of input variables, making it a critical technology for real-time risk management and hardware-accelerated machine learning.

Source Code
def f(x, z):
    a = x + z
    b = x * z
    return 0.5 * sin(a) * cosh(b)

print(f(1,2, 1,0))
Input Console
Input
Expressions
Adjoint mode
Output Console

Privacy Policy

Full Metal Analytics takes your data and privacy seriously.

We may process and store information that you provide to us as part of normal business operations.

By using our website or services, you hereby consent to our Privacy Policy and agree to its terms.

How we collect your data       The hosting provider for fullmetalanalytics.com, Amazon Web Services, collects your ip address for operational use and as part of its logging system for maintenance. In certain parts of fullmetalanalytics.com, you can enter information into forms to be submitted to Full Metal Analytics for processing. Any information provided in this way will be stored securely on the web server owned by Amazon Web Services until it is no longer needed at which point it will be removed.

What we do with your data       Any information provided to Full Metal Analytics will be processed by us and will not be shared with third parties without your explicit consent.

Version 3 - 16 Sep 2025