Recently, at Hot Chips 2023, Intel unveiled a groundbreaking experimental processor. This 8-core processor adopts some kind of RISC instruction set (possibly RISC-V). Its most eye-catching features are:
This processor fully demonstrates Intel’s powerful capabilities in optoelectronic integration. In fact, Intel has over 20 years of R&D and layout in silicon photonics technology. As early as 2004, Intel announced an optical modulator based on silicon wafers. At that time, other silicon optical modulators only had rates of 20MHz, but Intel’s product reached 1GHz in one go, representing a huge technological breakthrough.
In 2006, Intel announced the development of a laser chip integrating silicon and indium phosphide, which was faster than traditional copper wire transmission, laying the foundation for today’s silicon photonic components. In the more than ten years since then, Intel has continued to promote the development of silicon photonics technology and achieved commercialized mass production in 2016.
It can be seen that Intel’s technical accumulation and investment in the field of silicon photonics have become increasingly mature. This provides a solid technical foundation for Intel to develop optoelectronic integrated innovative processors.
Specifically, this new processor integrates 8 cores with silicon photonic components through Intel’s 2.5D packaging technology EMIB. Each core can connect quickly with other cores through optical interfaces, greatly improving the transmission speed between cores and nodes.
At the same time, using photons rather than electrons for signal transmission completely avoids the signal attenuation, heating and other difficulties faced in electronic transmission. This enables the construction of larger-scale parallel computing systems. In the past, “computing” and “communication” belonged to different industries, and optoelectronic integration technology is expected to promote the integration of the two.
In fact, Intel designed this processor in order to achieve more powerful parallel computing capabilities. Its goal is to provide support for DARPA’s image analysis processor project, HIVE, to achieve a speed 1000 times faster than existing technologies.
To this end, Intel adopted a massively parallel threaded design. A single core has 66 threads, and the 8 cores have a total of 528 threads. This design can give full play to the advantages of large-scale parallel processing in such workloads.
It can be foreseen that Intel’s innovative experimental processor has injected a shot in the arm for the future development of semiconductor technology. It not only demonstrates Intel’s outstanding capabilities in optoelectronic integration, but also defines new frameworks and blueprints for high-speed parallel computing.
In summary, Intel is still in the experimental stage. Theoretically, the HyperX topology network can be extended to over 100,000 OCP mezzanine cards and 1 million processors. A 16-processor OCP mezzanine card consumes only 1200W, which is an astonishing achievement. It is hard to imagine what it would be like if replaced with copper wires.
In the future, as processors evolve towards larger-scale parallelization and tighter optoelectronic integration, it will push the performance of computing systems to new heights. This will also inspire global semiconductor companies to conduct more in-depth R&D and layout in these cutting-edge technology fields. It is expected that Intel can continue to lead and play a leading role in the industry’s progress.
The following is Intel’s briefing: