Memory Interface Technologies

Innovations in memory infrastructure are poised to significantly improve the performance and cost effectiveness of mainstream data center applications. Business applications are under significant pressure to collect, analyze, store and deliver time-sensitive value to end customers while simultaneously driving down costs. These applications are drawing upon large quantities of data globally—sensor data, historical data, location data and customer-specific data—all of which must be scrutinized in real time. Memory infrastructure innovation is occurring along two vectors: near-memory innovations and far-memory innovations. 

Near-memory innovation improves business application performance by driving up the number of System-on-Chip (SoC) and Central Processing Unit (CPU) memory channels and improving the bandwidth of those individual memory channels while enabling innovative and lower-cost memory technologies to be connected seamlessly to the SoC. 

Far-memory innovation delivers shareable pools of memory resources that both drive up aggregate system memory utilization (and, therefore, drive down cost) and improve application performance as arrays of SoC devices can simultaneously operate on the shared memory pool. Memory infrastructure encompasses a broad range of new and emerging media types and interface technologies.

   

 

A broad range of silicon SoC devices, including mainstream CPU processors, application-specific Artificial Intelligence (AI) devices, network processors and other Application-Specific Standard Products (ASSPs) can all benefit from innovations in memory hierarchy. Multiple examples of new memory interface technologies, such as Open Memory Interface (OMI), Compute Express Link™ (CXL™) and Gen-Z have emerged. SoC devices with these memory interface technologies can be used to increase the number of independent memory channels while using far fewer SoC pins, reducing the cost of SoC packaging. These new interface technologies also are media independent. You can change the cost and performance profile of your deployed equipment by changing the type of media connected to these SoC interfaces. This type of SoC memory direct attachment is called near-memory technology.
     

Near-memory interface technologies can be used as on-ramp bridge interfaces to new memory interface technologies such as Gen-Z which includes routable fabric support. Gen-Z supports far-memory pools that can be shared at the rack level and beyond. Simultaneous and shared access to pools of far memory can dramatically improve the performance of data-centric computing, real-time analytics performance, High-Performance Compute (HPC) and machine learning applications at the server cluster scale all the way up to Warehouse Scale Computer (WSC) data center scale.
 

Major advances in memory infrastructure are being developed to meet the challenges of scaling memory bandwidth and availability to meet the demands of data-hungry applications. As a technology leader, we offer solutions like the SMC 1000 Smart Memory Controller that will help you keep up with these advances.