One of the biggest challenges for Artificial Intelligence and Machine Learning (AI/MI) is the movement of data from processor to processor (P2P) and processor to memory (P2M). The requirement for ever-increasing and complex models, is accentuating this problem and existing solutions are simply not fit for purpose, as they occupy too much real estate, consume too much power and limit bandwidth.
Avicena’s LightBundle™ solution solves these problems by offering ultra-low power, highly scalable interconnects using ultra-fast microLEDs.
This technology enables the efficient implementation of ever growing parameter sets in Large Language Models (LLMs).
Hyperscale datacenters continue to grow in number and size. While the longer reach interconnects up to 2km are well covered with conventional optical technologies, a paradigm shift is needed for shorter reach interconnects in emerging AI clusters where existing copper interconnects are running out of bandwidth and conventional laser based optical interconnects are too power hungry and costly.
This is where Avicena’s technology comes into play. By using LightBundle™ it is possible to provide a step improvement in power efficiency, bandwidth density, and scalability for interconnects; whether in the form of AOCs or chip-to-chip optical interconnects.
It should also be noted that ~75 percent of all data centre internet traffic happens within the datacenter, thereby putting increased pressure on local networking and interface bandwidth requirements. Avicena’s LightBundle™ technology offers a new level of efficiency in that data transfer, whether it is at the chip level or within the cabinets themselves.
Modern cars have a growing number of cameras and other sensors to support various safety devices and autonomous driving in the future. Some of these connections are starting to require fairly high bandwidth.
LED based links are ideally suited for automotive applications because the fiber optic connections are lighter than copper connections and LEDs can withstand the high temperatures typical in the automotive environment whereas lasers cannot.
The demand for high-speed wireless underwater communications is increasing, primarily due to the growth of off-shore wind turbines and the requirement to monitor their underwater structures.
Existing technologies fall short of the requirements for high-definition images and video requirements, necessary for this subsea monitoring. This has led to the development of optical solutions, using blue LEDs (blue light travels further in water than other wavelengths), but these existing systems are typically limited to data rates of < 10Mbps.
With Avicena’s LEDs it is possible to develop systems that can achieve data rates of greater than 10Gbps.
Ever more sensors and devices are now connected to the internet. This is also known as the Internet of Things (IoT).
E.g., modern camera sensors in smartphones are generating large amounts of data that need to be transported from the sensor to the processor in an efficient and non-intrusive fashion. Again, any electrical connection in a smartphone can act as an antenna or interfere with the rest of the circuitry. LED based interconnects do not suffer from this limitation, and they can tolerate the high temperature environment that some sensors operate in.
In 5G networks radio heads are getting ever more complex and consist of several different antennas all located in close proximity to each other. This requires a tight network of interconnects which is achieved mostly with electrical connections today.
The challenge is that any electrical connection in a radio head also acts as a potential antenna. There have been attempts at using other optical connections in the past, but laser efficiency starts to degrade at higher temperature, and past about 85°C laser failures increase significantly. LEDs can operate well beyond 125°C without major impact on efficiency or reliability.