Startup Code Metal Inc. launched today with a software platform designed to help developers optimize their code for edge computing devices.

The Boston-based startup has raised $16.5 million over two rounds to support its commercialization efforts. The bulk of the capital, $13.5 billion, came from a seed raise led by Shield Capital. The rest was provided in the form of a pre-seed round led by J2 Ventures.

“We are on the cusp of a massive transformation in software development, and edge computing stands to benefit the most from this change,” said Chief Executive Officer Peter Morales.

Processors don’t understand code written in Java, C++ and the other programming languages that are most commonly used to build enterprise software. As a result, applications written in those languages have to be translated into so-called machine code before they can be deployed on a server or edge computing system. Machine code is the low-level syntax in which processors express computations. 

Compilers, the tools that developers use to turn their software into machine code, not only perform translation but also optimize the applications they translate to make them faster. In some particularly complex software projects, developers perform parts of the optimization workflow manually to increase the performance gains. That task requires highly specialized skills and a significant time investment. 

Code Metal has developed a platform that promises to simplify the optimization phase of application projects. According to the company, its software places a particular emphasis on optimizing code for edge computing devices. Code Metal uses a combination of large language models and traditional code processing methods to make customers’ programs more efficient. 

One of the challenges that historically made optimizing software for edge systems difficult is that such devices often have limited processing power. In many cases, the amount of memory at their disposal is constrained as well. Those factors limit the amount of data that an application can process at once. 

Many edge systems are battery-powered, which means developers also have to take energy usage into account. Programs must carry out processing in a way that leaves enough electricity for other tasks. 

One of the ways that developers optimize programs, particularly AI applications, is by fine-tuning the way they perform parallelization. That’s the task of splitting up a complex computation into smaller calculations and running each calculation on a different set of circuits within an edge device’s chip. Performing calculations side-by-side on different circuits rather than one after another speeds up processing.

Parallelization is not as simple as assigning each part of a computing task to a different set of circuits. In many cases, there’s no way to run computations side-by-side simultaneously because one calculation can only begin after the results of another become available. Developers have to account for such constraints while still parallelizing as many of an application’s calculations as possible.

Code Metal says that its platform can optimize applications written in a variety of programming languages. It supports Python, which is widely used for AI development, as well as more specialized syntaxes such as Julia and Matlab. The latter two languages are mainly used for scientific tasks such as running physics simulations.

Developers can use Code Metal to not only optimize their code for edge devices but also debug it. The platform provides the ability to create unit tests, scripts that automatically scan an application for errors one small block of code at a time. Code Metal can also perform regression testing, or the task of checking a program for errors after every single code change. 

The company says that its platform significantly reduces the amount of effort required to optimize applications for edge computing chips. Depending on the project, the software promises to cut development times by between 50% and 75%.

Code Metal will use its newly raised capital to hire more workers. According to the company, the recruiting push will focus on hiring engineers with AI, compiler design and hardware expertise.

Photo: Unsplash