One of the most dynamic applications for memory technologies is that of server configurations. Driven by the massive amount of data being created and transported, servers have become the application known for design innovation and pushing the performance envelope. Enterprise SSDs based on NAND technology have clearly established a high volume niche between the traditional standbys of high performance DRAMs and lower cost per bit of hard disk drives.
However, we have felt that there was yet more to come from this application segment. The cost/performance pressures were increasing in a way that was creating a wider range of possible performance attributes rather than simply driving the cost downward in a relatively commoditized market. We believed that this application would be one of the primary market entry points for the range of new and emerging memory technologies.
A new paper from Chuo University in Tokyo supports describes a simulation of a hybrid SSD that contains 1Gbyte of non-volatile Resistive RAM as a buffer to a traditional 256 GByte of NAND storage capacity. The paper noted that the resulting performance would achieve switching times of 10ns (compared to 100,000ns of existing NAND SSDs), while still retaining the power consumption advantages of nonvolatile memory technologies over DRAMs. The controller algorithms could then store frequently-used data in the resistive RAM cache from which it could be gathered more quickly than from NAND, while the resistive RAM cache was large enough so that the data held there could be retained until the was a sufficient quantity to efficiently write it out to the NAND cells. This configuration eliminated much of the small random write sequences and data fragmentation issues inherent in the normal usage of NAND.
Additionally, due to the reduced number of erase/re-write sequences, the useful life of the NAND devices could be extended to more than seven times greater than in existing SSDs.
The team also concluded that up to 79% energy savings could be achieved due to a combination of the reduced number of erase/re-write cycles and anticipated advances in Thru Silicon Vias to support packaging configurations containing the controller, resistive RAM, and NAND in the same package.
We believe that the question of when the cost per bit of the new memory technologies will compete with DRAM or NAND is no longer a useful measurement. Following the example of system performance improvements based on integrating NAND SSD into Server applications, new memory technologies are not directly replacing older technologies but will enable new performance possibilities. The key question is when does the cost/performance of the new memory technologies trigger the next wave of Server re-designs—and that question should be posed to the OEM system designers. This issue and others are addressed in our upcoming report on Server Applications as the Market Entry Point for new and emerging memory technologies.
