The rise of AI chips (UPSC Current affairs)

Relevance: GS3 - Awareness in the fields of IT, Space, Computers, Robotics, Nano-technology, Bio-technology and issues relating to Intellectual Property Rights.

Click here for Daily Current Affairs

Why in the News?

• Artificial Intelligence (AI) chips are becoming more widely used, with chipmakers developing several types to support AI applications such as natural language processing (NLP), computer vision, robotics, and network security across a number of industries, including automotive, IT, healthcare, and retail.
• Nvidia, the market leader, recently unveiled the H100 GPU, which is believed to be one of the world's largest and most powerful AI accelerators, with 80 billion transistors.
• Nvidia's competitor Intel introduced new AI processors earlier this month to give clients with deep learning computing options for training and inference in data centres.
• One of the primary reasons driving the market's growth is the rising deployment of AI chips in data centres.

What exactly are AI chips?

• To enable deep learning-based applications, AI chips are created with a specialised architecture and integrated AI acceleration.
• Deep learning, also known as active neural network (ANN) or deep neural network (DNN), is a type of machine learning that falls under the umbrella of artificial intelligence.
  • It is made up of a set of computer commands or algorithms that stimulate brain activity and structure.
  • DNNs go through a training phase when they gain new skills by analysing previous data.
  • DNNs may then generate predictions against previously unknown data by employing the skills learnt during deep learning training.
  • Deep learning may speed up and simplify the process of gathering, analysing, and interpreting huge quantities of data.
• With various hardware designs and complementing packaging, memory, storage, and connectivity technologies — these chips enable AI to be infused into a wide range of applications to assist in the transformation of data into information and ultimately into knowledge.
  • AI chips, such as application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), central processing units (CPUs), and graphics processing units (GPUs), are intended for a variety of AI applications.

How do they differ from regular chips?

• When conventional chips containing processor cores and memory perform computational tasks, commands and data are constantly transferred between the two hardware components.
  • These processors, on the other hand, aren't appropriate for AI applications since they can't manage the greater computing demands of AI workloads with large amounts of data.
  • However, certain higher-end conventional CPUs may be capable of processing AI applications.
• AI chips often have processor cores as well as one or more AI-optimized cores (depending on the chip's size) that are meant to work together while executing computational tasks.
  • Due to close interaction with the other processor cores, which are designed to handle non-AI applications, the AI cores are optimised for the needs of heterogeneous enterprise-class AI workloads with low-latency inferencing.
• AI chips, allows smart devices to execute advanced deep learning tasks such as object identification and segmentation in real time while consuming little power.

What applications do they have?

• Semiconductor companies have created a variety of specialised AI chips for a variety of smart machines and gadgets, including ones that claim to bring data center-class performance to edge devices.
• Some of these chips enable in-vehicle computers to execute cutting-edge AI applications faster.
  • Wearable gadgets, drones, and robots all use AI processors to enable computational imaging applications.
• Additionally, the need for chatbots and online channels such as Messenger, Slack, and others has expanded the use of AI chips for NLP applications.
  • They analyse user communications and conversational reasoning with NLP.
• Then there are chipmakers that have created AI processors with on-chip hardware acceleration, aimed at assisting clients in gaining business insights at scale across banking, finance, trading, insurance, and consumer interactions.

What can we anticipate for the future?

• Cerebras Systems' brain-scale AI solution set a new benchmark, opening the path for more sophisticated systems in the future.
  • Its CS-2 is a single wafer-scale semiconductor with 2.6 trillion transistors and 8,50,000 AI optimised cores, driven by the Wafer Scale Engine (WSE-2).
  • According to the company, the human brain has on the order of 100 trillion synapses, and a single CS-2 accelerator can handle models with over 120 trillion parameters (synapse equivalents).
• Neuromorphic computing, another AI chip design technique, employs an engineering process based on biological brain activity.
  • The utilisation of neuromorphic chips in the automobile sector is predicted to expand in the next few years.
• The rise in demand for smart homes and cities, as well as a spike in investments in AI start-ups, is likely to propel the worldwide AI chip market forward.
• From 2021 through 2030, the global AI chip market is predicted to develop at a compound annual growth rate (CAGR) of 37.4 percent, from $8.02 billion in 2020 to $194.9 billion in 2030.