BoolSi Raises $6M in Seed Funding Led by Fine Structure Ventures

The round also saw participation from Pillar VC, Fifth Quarter Ventures, and Coalition Ventures.

The company plans to use the funds to grow its operations and continue developing its technology.

Field-programmable gate arrays (FPGAs) can be reconfigured after manufacturing and placed alongside a central processing unit.

Read More:Pramaana Labs Raises $27M in Seed Funding Round

On the user side, the process is simple. A developer points the compiler at a slow part of a program written in C, C++, or another high-level language, and BoolSi generates a custom circuit along with a driver.

Custom hardware can run specific workloads much faster than a general purpose CPU because it uses dedicated circuits that operate in parallel avoiding the repeated instruction processing overhead of a CPU.

However, the main challenge has always been complexity. Chip design is extremely detailed and requires coordinating millions of tiny components, which makes it more like watchmaking than traditional software development. There are few open-source tools, high barriers to entry, and it often takes years to learn. As a result, most software engineers never move into hardware, even when their applications could benefit from it.

BoolSi changes the problem from translating code into hardware to understanding what the program actually does. It trains machine learning models that turn software into digital circuits, using the program itself to generate synthetic training data.

A fuzzer explores different inputs and converts every run into a perfectly labeled example, which the company says makes 100% accuracy both possible and necessary. For verification, BoolSi trains multiple independent models in parallel and checks them against each other.

The company highlights its performance using a simple benchmark: a regex task that finds email addresses in text. When compiled with gcc -O3 on an ARM Cortex-A9 processor, it takes 2.66 milliseconds. A single BoolSi hardware agent completes the same task in 0.325 milliseconds, about 8x faster.

When eight agents run in parallel, the task finishes in just 0.042 milliseconds, making it about 63x faster than the CPU. BoolSi compares its approach to tools like Vivado HLS, Catapult, and Bambu, saying those tools improved productivity for hardware engineers but did not make hardware design accessible to software developers.

The startup is initially targeting embedded developers in robotics, where tasks like motor control, sensor fusion, and predictive control put heavy strain on general-purpose CPUs.

A private beta is planned for the third quarter. BoolSi is starting with FPGAs because they allow developers to deploy quickly, and the company expects its toolchain to later expand to custom ASIC chips as workloads become more stable.

About BoolSi 

Founded in 2023 by Mihailo Isakov, BoolSi is a Boston-based startup that builds a platform to convert regular software code into custom silicon hardware accelerators automatically.

Read More:Ralo Raises $2.9M in Seed Funding from Y Combinator