Moore's law is a prediction made by the president of Intel in 1965, Gordon Moore, who predicted that the number of transistors per circuit would double every 18 to 24 months. This trend was a supported thanks to transistor scaling technology. where smaller transistors allowed to reduce the voltage, what enabled an increase in CPU frequency as well known as the Dennard scaling. Moore's law also brought cost scaling since it duplicated number of transistor the cost per transistor halved also consume less energy Moore's law is ending now, but Dennard scale already ended when the voltage reached a threshold limit. Since Dennard scaling stopped in 2010, the way to achieve more performance per chip is to introduce more parallelism, accelerators, and and an increased number of cores per chip. However, this introduced several challenges: how to program it. Hardware accelerators, and further parallelism greatly benefits applications which are compute-bound However, other applications which are memory bound don't scale well with the number of cores in current memory hierarchies. For those applications I'm researching new programming models that execute instructions where the data is local, as opposed to bring in the data to the cores. This new programming model, where the memory is more programmable or exposed the software allows the compiler or the programmer to decide where the data is located. I also created a memory access parallel load engine, where the data movement can be programmed explicitly, so that irregular memory access patterns that don't benefit from caches can be improved and specialized.