Artificial Intelligence: Impact on the Connector Industry

Computer intensive applications have stimulated the growth of large-scale specialized AI infrastructure. Due to the demand for superior speed, reduced latency, and energy efficiency, the connector industry will benefit from the development of AI.

Artificial intelligence quickly entered the public eye with the release of ChatGPT on November 30, 2022, and its influence has grown exponentially since then. This transformative technology has been implemented in many applications and has brought record growth to key component suppliers such as Nvidia. The ultimate development of artificial intelligence is still in its very early stages, but as far as I know, it has the potential to ultimately impact almost every aspect of our lives.

Computers that support AI can create unique information, correlate large amounts of data, and use processes such as human intelligence to perform tasks. By utilizing information from a wide range of training processes, AI computers can analyze data to create new innovative solutions. Imagine a system reviewing clinical reports of thousands of patients to compare the success of their treatments with the success of their outcomes, in order to determine the most effective treatment for individuals with similar symptoms, age, gender, genetic, and racial backgrounds. In another example, medical researchers can insert the desired properties of organic materials, and AI will search its training database for molecular structures of materials with potential therapeutic effects on the human body. Customized algorithms and training of artificial intelligence computers can review every financial transaction of a specific stock and the current economic situation to identify trends that can strongly predict the future performance of that stock. AI is being injected into robots that can learn new tasks by simply observing humans, eliminating expensive and time-consuming programming.

AI systems can not only collect and analyze data, but also understand the context of requests. In a recent demonstration, a human told a robot to 'clean this table' with a series of plates, cups, and glasses on it. The robot understood the intention of the request and continued to gently place each item on the shelf, cleaning the surface. When asked about the robot's feelings about completing tasks, it said, "I think I do a great job. These are not programmed. Chatbots are getting better at understanding spoken language and speaking in natural language. Chat GPT is trained to understand what makes things humorous and create jokes that are more interesting than those written by human professionals. These systems are currently far from perfect, but are making significant progress towards general artificial intelligence. The ultimate step will be a system with self-awareness and self training at a speed far beyond human capabilities - this is a terrifying concept.

Due to the demand for superior speed, reduced latency, and energy efficiency, the connector industry will benefit from the development of AI. Computer intensive applications have stimulated the growth of large-scale, specialized AI infrastructure. High speed backplane, mid board, orthogonal connectors, and I/O connectors will be required, but equipment sockets and high-density, flat cable to PCB connectors will also be needed. Pluggable optical transceivers including QSFP-DD, OSFP, and OSFP-XD are rapidly being adopted to provide high bandwidth interconnection between remote AI clusters. The advanced versions of these transceivers use coherent and linear technologies to meet transmission distance requirements of thousands of kilometers while still reducing power consumption.

We are considering a new system architecture, such as co packaging optical components and chicks. The current trend in decomposing systems is that critical resources (including memory, storage, and computing) may reside in remote locations, which will require high-performance, low latency optical interconnects. Distributed computing resources also disperse the demand for local electricity. New copper cables and fiber optic interconnects with necessary density and performance are needed to support the unique requirements of AI computing clusters.

The current electricity consumption of AI computers is increasing at an unsustainable rate. A single rack can consume 10000 watts of power. We will need connectors designed to minimize resistance losses to the greatest extent possible. The current fan driven air cooling will be replaced by liquid cooling. Initially, only selected devices including processors will be cooled through cold plate technology, but over time, more and more devices will require liquid cooling, and eventually full immersion cooling technology may be adopted. All types of connectors used in these systems need to be modified to be compatible with these cooling technologies.

The communication between high-performance servers, routers, and switch racks will require the use of high bandwidth copper and fiber optic cables. AI is seen as a technology that will drive the use of optical devices within data centers, boxes, and closer to switches. Low cost direct connection passive copper cables can meet the needs of shorter rack cables, but with increasing distance and bandwidth, the use of active copper cables, active optical cables, pure optical cables, and transceivers will increase.

A few years ago, some equipment manufacturers believed that using component design systems defined in industry standards would suppress their innovation capabilities. Standards take several years to be established, and upgrades require even longer time. In some cases, manufacturers are willing to specify proprietary components to gain a competitive advantage. The result often leads to higher costs and supply issues. Open standards have become ubiquitous as manufacturers require at least two sources, and many component suppliers have accepted product tools defined by standards. Samtec has been very successful in adopting this strategy.

High speed interconnection (such as Samtec's active optical component series) supports the development of artificial intelligence and other advanced technologies

Processing new high-performance connectors has become very expensive, especially when the available market may be limited. Standard organizations such as Ethernet, CXL, and PCIe have recognized the challenge of maintaining relevance and have been actively upgrading their standards to meet the demand for improved performance. The Ethernet Alliance is particularly active in releasing the 1.6Tb Ethernet roadmap. Standard organizations see real opportunities in emerging AI systems.

Ultimately, artificial intelligence is becoming a powerful new technology that will drive the entire electronics industry to new levels of performance. Skyquest Technology recently estimated that the global value of hardware designed specifically for AI based applications will be $13.3 billion in 2022, and is expected to reach $84.9 billion by 2031. The value of standard hardware will be many times greater than it.

To maintain the astonishing pace of AI progress, it is necessary to develop advanced interconnections within an accelerated design cycle. The popularization of AI will provide excellent growth opportunities for component manufacturers who keep up with industry demands and are committed to developing new products that support this emerging demand.