The technological direction and value of future memory device required for building future AI application environments and systems can be predicted as follows:

1. Scaling & Stack-up Value

It refers to the technological value and vision created based on the achievement of continuous memory bit growth by overcoming the scaling limits of existing DRAM and NAND devices. It is a value as a memory capacity required to operate AI models using a huge amount of data, and the key factor for the creation of economic value through the increase of bits / cost and bits / area. In order to achieve the values, in the case of DRAM, technological innovation for 2D scaling extension and 3D platform era should be followed up. In the case of NAND, 3D platform technology has already been commercialized, and it is expected that technological innovations will be attempted for increased integration and expanded applications.

1. Hybrid Emerging Value

This is the value of creating a new layer of memory device opportunities and new application fields through the use of characteristics that existing DRAM and NAND devices do not have. This is the value that can be applied to application environments and computing structures that require large capacity, low cost, and non-volatile characteristics compared to DRAM, or fast speed and long operating life compared to NAND. To this end, continuous discovery of various types of memory device technologies and preemptive research activities are important from a mid- to long-term perspective. In addition, in order to create new business value, the expansion of application environments and systems that require Hybrid Memory devices must be accompanied. Through this, the future market size, application timing, etc. are expected to be determined, and based on this, the research and development strategy is also expected to be more specific.

1. Convergence Value & Beyond

The computing structure is expected to further evolve into a heterogeneous computing environment consisting of CPUs, GPUs, and various types of AI accelerators. In this heterogeneous convergence environment, a system configuration suitable for creating power, performance, area, and cost-effective values related to the movement of data is more important. In addition to the existing scaling approach, the provision of technological solutions through heterogeneous integration from a system perspective is expected to increase for logic and memory devices. This will enable the expansion of the application areas of future memory device technology, and new opportunities and values are expected to be created. In other words, not only the storage capacity of data, but also the bandwidth (Bits / sec) and movement energy (J / bit) related to data movement are expected to be important business judgment values. Therefore, in the future, along with low-power, high-performance, and high-bandwidth memory technology innovation, technology leadership for building heterogeneous convergence systems in 2.5D and 3D forms of logic and memory devices, and cooperative development strategies for this are expected to become more important.

Lastly, it seems that technological preparation for the Non Von Neumann environment where the boundaries between memory and computational functions disappear and are integrated is necessary. Starting with the technological challenge of including AI computational units in memory chips where data is stored to overcome the fundamental limitations of data movement in Von Neumann-based computing structures, it is expected that technology will evolve in the direction of simultaneous computations within the memory cell matrix. This technological innovation is the starting point of neuromorphic semiconductor technology that best mimics the structure and behavior of human neural networks, and will be a new opportunity and innovative challenge for future memory technology. Research and development strategies to prepare for this are also very important, and it is expected that strengthening integrated (end to end) research capabilities that consider the convergence of memory and logic technologies, materials-processes-devices-design-architecture-S/W-workload, etc. from the beginning and building a collaboration ecosystem for this will become even more important.