ZeroPoint Raises $5.5M for On-Chip Data Compression Technology

ZeroPoint Technologies AB, a startup with hardware that helps processors use DRAM more efficiently, has secured $5.5 million in early-stage funding.

In announcing its Series A funding today, the company said Munich-based Matterwave Ventures was the lead investor, with participation from Industrifonden, Climentum Capital and Chalmers Ventures.

ZeroPoint is a spin-off from Chalmers University of Technology in Gothenburg, Sweden. The startup was formed to commercialize research into data compression techniques that co-founders Per Stenström and Angelos Arelakis conducted at the university. ZeroPoint says its technology can make data center servers more power efficient, reducing the cost of running those servers in the process.

The speed at which a server processor can fetch data from DRAM directly impacts its performance. As a result, reducing data acquisition times has become a top priority for semiconductor designers. One way to speed up the acquisition workflow is to compress the files that the chip processes. The smaller the file, the faster it can be pulled from memory.

Using compression to speed up chips in this way has historically been impractical because most advanced compression algorithms are relatively slow, negating the processing speedups that can be achieved by optimizing DRAM usage. Faster algorithms exist, but they are not as good at reducing data.

ZeroPoint claims to have solved this trade-off. The company offers a small semiconductor module “Ziptilion-BW”, which is said to have excellent data compression and high speed. The chip promises to reduce memory usage for datasets by up to 400% while running 1,000 times faster than some traditional compression methods.

Ziptilion-BW's performance is made possible by what ZeroPoint describes as a set of entropy-based compression algorithms. The company says that these algorithms use statistical techniques to analyze the chip's memory to determine the type of data it contains. Ziptilion-BW then uses this information to find the most efficient way to compress the data.

The chip combines ZeroPoint's entropy-based algorithms with many other performance optimization techniques, notably storing recently used information in an on-board cache so that the server can retrieve information from this cache more quickly than from the DRAM where that data is typically stored, accelerating calculations.

According to ZeroPoint, Ziptilion-BW can be implemented as a semiconductor module measuring just 0.4 square millimeters, making it relatively easy for chipmakers to integrate the technology into their processors. ZeroPoint estimates that Ziptilion-BW can improve data center servers' performance per watt by 50%, reducing total cost of ownership by 25% in the process.

“ZeroPoint Technologies' customers include the world's largest semiconductor companies, and our products are in demand with data center operators looking to overcome the mounting challenges of memory bottlenecks,” said ZeroPoint CEO Klas Moreau (pictured, center, and co-founders Stenstrom and Arellakis).

The company plans to use the proceeds from the newly announced funding round to launch additional hardware accelerated products. ZeroPoint also plans to expand its sales efforts and hire more employees in the United States and Sweden to support its growth initiatives.

Photo: Zero Point