High speed PWM DAC design for Analog In Memory Computing Mixed Signal IP

Artificial intelligence is revolutionizing numerous sectors, requiring increasingly efficient and high-performance computing systems. This thesis, carried out at STMicroelectronics, explores the implementation of various digital-to-analog converter architectures within Analog-in-Memory Computing systems, aiming to optimize power consumption and the occupied area of the system. The thesis project is structured into several sections. Initially, the theoretical foundations of artificial intelligence are introduced, with particular attention to computational requirements and the current challenges related to the large number of computations needed. Subsequently, the general architecture of a system based on Analog-in-Memory Computing is defined, highlighting its main advantages in the context of neural networks. The central part of this work focuses on the design and architectural choices of the converters, analyzing their key characteristics. The results obtained show that certain conversion architectures offer significant advantages in terms of power consumption and area minimization without compromising the operating frequency. In particular, the use of optimization techniques has allowed for an optimal balance between energy efficiency and performance. The conclusions of this thesis project provide new perspectives for the design of circuits intended for Analog-in-Memory Computing, suggesting future directions for further improvements and practical applications, paving the way for more efficient and scalable solutions for artificial intelligence.