EdgeCortix’s MERA Compiler and Software Framework serves as a comprehensive platform for deploying neural network models across various systems. It facilitates AI inference by combining the Dynamic Neural Accelerator (DNA) technology, offering tools like APIs, a code generator, and runtime capabilities essential for implementing pre-trained deep neural networks. The framework is engineered to tackle innovative generative AI applications, accommodating a spectrum of leading processors, such as AMD, Intel, Arm, and RISC-V, ensuring speedy integration into existing infrastructures.
MERA stands out by supporting an array of ML frameworks, including TensorFlow Lite and ONNX, and offering native interfaces for Python and C++ to customize AI workflows. It simplifies the development and deployment process with post-training model calibration and quantization. The inclusion of Apache TVM and MLIR adds flexibility, allowing developers to advance from modeling to deployment with ease.
The framework is integral to the company's suite of offerings as it optimizes neural network computation, resource allocation, and execution order, all within a runtime reconfigurable architecture. The MERA platform thus grants software developers the ability to focus on AI model refinement without getting entangled with hardware intricacies, dramatically reducing latency and boosting throughput efficiency.
