The company is working on a processor-in-memory (PIM) architecture for use in the data center where it would sit embedded in DRAM to be close to data. By reducing the need to move data, the use of hundreds or thousands of processing elements in DRAM working to the main CPU as a controller, could increase performance by factors of 10 to 25 compared with traditional server architectures with close to no additional power consumption, Upmem has claimed.
"Initial benchmarks by our partners validate the game-changing added-value of Upmem PIM technology, as well as the strong fit of its programming model for a large scope of real world data-intensive applications," said Gilles Hamou, CEO and co-founder of Upmem, in a statement.
It remains unclear who those partners are and whether they are working totally in simulation or with FPGA prototypes of an architecture that will eventually be built in a leading-edge DRAM manufacturing process.
The PIM chip integrates a proprietary RISC processors known as DRAM processing units (DPUs) and main memory. It comes with a software development kit that can be used to accelerate data-intencive applications running in the data center.
"Faster and more efficient data analytics require new datacentric application architectures, positioning compute nearer the data," said Western Digital iMemory project leader Robin O’Neill, in the same statement. "The Upmem processing-in-memory solution is particularly relevant and highly promising for a variety of data analytics use cases, without dramatic changes to server architectures."
Upmem said the financing round will enable the company to produce and bring to market its chip-based solution. In parallel, Upmem will accelerate its evaluation programs with top tier global big data customers and IT labs, using available programming and simulation tools.
Related links and articles: