On-die cache memory is a key component in advanced processors since it can boost micro-architectural level performance at a moderate power penalty. Demand for denser memories only going to increase as the number of cores in a microprocessor goes up with technology scaling. A commensurate increase in the amount of cache memory is needed to fully utilize the larger and more powerful processing units. 6T SRAMs have been the embedded memory of choice for modern microprocessors due to their logic compatibility, high speed, and refresh-free operation. However, the relatively large cell size and conflicting requirements for read and write make aggressive scaling of 6T SRAMs challenging in sub-22 nm. In this dissertation, circuit techniques and simulation methodologies are presented to demonstrate the potential of alternative options such as gain cell eDRAMs and spin-torque-transfer magnetic RAMs (STT-MRAMs) for high density embedded memories.Three unique test chip designs are presented to enhance the retention time and access speed of gain cell eDRAMs. Proposed bit-cells utilize preferential boostings, beneficial couplings, and aggregated cell leakages for expanding signal window between data `1' and `0'. The design space of power-delay product can be further enhanced with various assist schemes that harness the innate properties of gain cell eDRAMs. Experimental results from the test chips demonstrate that the proposed gain cell eDRAMs achieve overall faster system performances and lower static power dissipations than SRAMs in a generic 65 nm low-power (LP) CMOS process. A magnetic tunnel junction (MTJ) scaling scenario and an efficient HSPICE simulation methodology are proposed for exploring the scalability of STT-MRAMs under variation effects from 65 nm to 8 nm. A constant JC0*RA/VDD scaling method is adopted to achieve optimal read and write performances of STT-MRAMs and thermal stabilities for a 10 year retention are achieved by adjusting free layer thicknesses as well as projecting crystalline anisotropy improvements. Studies based on the proposed methodology show that in-plane STT-MRAM will outperform SRAM from 15 nm node, while its perpendicular counterpart requires further innovations in MTJ material properties in order to overcome the poor write performance from 22 nm node.