All articles

The MLPerf™ Tiny benchmark are a representative set of deep neural nets to assess the performance of embedded devices on Artificial Intelligence (AI) workloads. This application note demonstrates how ComputeRAM™ dramatically enhances an embedded microcontroller's energy efficiency and throughput while retaining the flexibility to process different deep neural network architectures.

Matrix-vector multiplications (MVMs) are fundamental operations in linear algebra, crucial for scientific computing workloads ranging from AI to signal processing. This application note describes how, by simply replacing the on-chip memory of a reference MCU (microcontroller unit) with Synthara's ComputeRAM™, up to 139x faster speeds and 158x better energy efficiency can be unlocked.

In-memory computing (IMC) integrates processing within memory to enhance speed and efficiency for AI and other data-intensive edge applications, but faces challenges like complex integration and software demands. Synthara’s ComputeRAM™ capitalizes on IMC's advantages while addressing all its limitations, underscoring the need to understand this transformative technology.