Microsoft Research’s cover photo
Microsoft Research

Microsoft Research

Think Tanks

Redmond, Washington 382,611 followers

We advance science and technology to benefit humanity.

About us

At Microsoft Research, we accelerate scientific discovery and technology innovation to empower every person and organization on the planet to achieve more. We do this by bringing together the best minds across diverse disciplines and backgrounds to take on the most pressing research challenges for Microsoft and for society. Our Research Lens We consider research directions through the lens of the positive impact we aspire to create with and for customers, communities, and all of society.

Website
http://www.microsoft.com/research
Industry
Think Tanks
Company size
1,001-5,000 employees
Headquarters
Redmond, Washington
Founded
1991

Updates

  • Microsoft Research reposted this

    View organization page for Microsoft

    27,859,904 followers

    Think of datacenter networking like water moving through pipes. Traditional optical cables move data through a narrow, fast stream of light. While Microsoft’s new MicroLED system sends data through thousands of channels at once – more like a wide, slow-moving river carrying the same volume. The result: a new approach that researchers estimate could use about 50% less energy than mainstream laser-based optical cables, while supporting the massive data flows behind AI and cloud services. Learn how: https://msft.it/6001QoewP

    • No alternative text description for this image
  • Microsoft Research reposted this

    Today’s announcement from Microsoft, based on technology invented by Microsoft Research, is poised to advance the way AI (and other) systems are built. The innovation uses simple LED chips to transmit data, instead of copper wires or fiber-optic (laser-based) communication. https://aka.ms/AA101ymw The commercialization of this invention will make our AI datacenters consume less energy, be more reliable, and open additional design options to make our systems much better. Microsoft decided to share this invention with the world and make it widely available, as lower energy consumption across the industry is good for all of us. Sending data along copper cables using electric pulses is standard because copper is cheap. As you increase data transmission rates, the copper wires must get shorter. That limitation has forced us to place AI and other chips ever closer together, creating a rapid increase in a rack’s power density, and making the challenges of cooling the rack much harder. That trend is unsustainable. The other alternative today is the use of lasers over fiber optic cables. Fiber optics can go long distances and transmit data at high bandwidth. The challenge with using them is that the laser-based transmitters are bulky, expensive, have relatively high failure rates, and consume much more power than copper. A Microsoft Research team in our Cambridge Lab had an out-of-the-box idea to address this problem in 2020: Use commercial LED chips (like we have in our cell phones) to transmit thousands of parallel streams of data at low power over low-cost medical imaging fiber. This breakthrough will allow the computing industry to transmit data with less power—at high bandwidth—than either copper or fiber optics, up to distances of tens of meters, at potentially higher reliability. The team solved many problems along the way, getting the LEDs to cycle at high speeds while remaining power efficient, finding a way to transmit those many streams of light reliably, coupling the light into the imaging fiber (since LEDs spread light in all directions), and keeping dispersion of the light sufficiently low that it can work over the target distances reliably. They formed a deep partnership with product engineers in Microsoft’s Azure Core, Azure Hardware, and M365 divisions and external partners (MediaTek and others). Together with their partners, they built a proof-of-concept solution that is plug-compatible with today’s datacenter cables. Microsoft Research teams are free to explore any ideas or fundamental research. When they produce a breakthrough, we have decades of experience getting new technology to world-wide scale. But it takes grit, and I’m proud of the perseverance of the team. Thanks to Ariel Gomez Diaz, Junyi Liu, Kai Shi, Kaoutar Benyahya, Marianna Pantouvaki, Shawn Yohanes Siew, Doug Kelly, Daniel Cletheroe ,Vassily Lyutsarev, Brian Robertson, Thomas Burridge, and Paolo Costa--and all of their collaborators--for their hard work.

  • In this issue: Phi-4-reasoning-vision highlights advances in compact multimodal reasoning models, new systems improve LLM inference serving and benchmark AI agents in real-world network environments; CineScene advances cinematic video generation with consistent scenes and camera motion; SibylSense introduces adaptive rubric learning for open-ended LLM tasks; and new research explores AI’s role in improving individual and population health.

  • View organization page for Microsoft Research

    382,611 followers

    The latest Microsoft Research Forum episode is now available. • ARO: A new lens on matrix optimization for LLMs (Chao Ma and Wenbo Gong) • Dion2: A new simple method to shrink matrix in Muon • Magentic Marketplace: Testing societies of agents at scale (Gagan Bansal) • Agent Lightning: One learning system that makes all agents evolve (Luna Qiu) • Teaching small language models to think like optimization experts with Optimind (Xinzhi Zhang) • Lessons from deploying HealthBots with experts-in-the-loop (Mohit Jain) Register to watch: https://msft.it/6040Q5fT8

Affiliated pages

Similar pages