 |
 |
 |
Progressively, operations between scalars (numbers) were parallelized into operations between vectors and, subsequently, matrices. Multiplication between matrices now trends as the most time- and energy- demanding operation of contemporary AI computational systems.
A technique called "tiled matrix multiplication" (TMM) helps to speed computation by decomposing matrix operations into smaller tiles to be computed by the same system in consecutive time slots. But modern electronic AI engines, employing transistors, are approaching their intrinsic limits and can hardly compute at clock-frequencies higher than ~2 GHz.
The compelling credentials of light - ultrahigh speeds and significant energy and footprint savings - offer a solution. Recently a team of photonic researchers of the WinPhos Research group, led by Prof. Nikos Pleros, from the Aristotle University of Thessaloniki, harnessed the power of light to develop a compact silicon photonic compute engine capable of computing TMMs at a record-high 50 GHz clock frequency.
As reported in Advanced Photonics, they employ silicon-germanium electro-absorption modulators and a novel neuromorphic architectural design capable of encoding and computing data. According to corresponding author George Giamougiannis, "This work paves the way for the resolution of DL-based applications that require line-rate computations," and the work promises to contribute significantly to data center cybersecurity.
Data center cybersecurity: Light hunting the evil
Undoubtedly, the AI burst has equipped both benign and wicked users with strong toolkits to speed-up and automate their activities. With the data travelling in the data centers (DCs) augmenting by ~13 percent year-by-year, they have become a major target for malicious individuals who aim to compromise sensitive data, e.g., financial data, personal information, and intellectual property of many organizations, including government agencies, military forces, hospitals, and financial institutions. On that account, DC cybersecurity is imperative to prevent invaders accessing classified information.
Indeed, threat detection mechanisms face a new set of requirements resulting from the quantity of data flowing through the vast number of servers and switches within contemporary DCs. Real-time threat detection is imperative: packet inspection must be processed at ultrahigh speeds. Moreover, threats must be detected as early as possible within the route of the malicious packets: every DC node should be equipped with a powerful cybersecurity toolkit.
Exploiting their ultrafast processor, the researchers from Aristotle University of Thessaloniki, in collaboration with NVIDIA's experts in the field of DC cybersecurity, successfully merged silicon photonics with AI to establish a framework to identify successfully and swiftly one of the most common types of DC attacks, namely distributed denial-of-service (DDoS) attacks, in NVIDIA's servers at line-rates. Thanks to this novel computational scheme, the number may soon be up for DC attacks - at least for the time being.
Research Report:Matrix multiplications at the speed of light
Related Links
International Society for Optics and Photonics
Space Technology News - Applications and Research
| Subscribe Free To Our Daily Newsletters |
| Subscribe Free To Our Daily Newsletters |
