IBM unveils AI processor chip to help financiers deal with fraud

The AI processor chip will help businesses move towards a fraud prevention posture from a fraud detection posture.
24 August 2021

The AI processor chip will help businesses move towards a fraud prevention posture from a fraud detection posture. Source: IBM

  • New chip design unlocks the ability to leverage deep learning inferencing on high-value transactions
  • Designed to greatly improve the ability to intercept fraud, among other use cases
  • Fraud cases continue to increase despite increasing AI-based fraud detection tools

The IBM Telum Processor is IBM’s first AI processor chip. Designed to bring deep learning inference to enterprise workloads, especially to address fraud in real-time, the chip contains on-chip acceleration for AI inferencing while transactions take place.

Announced at the Hot Chips Conference, the Telum Processor was developed over three years and is expected to revolutionize fraud detection in banking, finance, trade, insurance applications, and customer interactions. A Telum-based system is planned for the first half of 2022.

Highly regulated industries like banking and finance may find IBM Telum fitting in perfectly for them, as the chip is designed to enable applications to run efficiently where the data resides. This overcomes traditional enterprise AI approaches that tend to require significant memory and data movement capabilities to handle inferencing.

With AI being embedded in almost everything an organization does today, recent IBM research showed that 90% of respondents felt that being able to build and run AI projects wherever their data resides is important.

The AI processor chip will help businesses move towards a fraud prevention posture from a fraud detection posture

The AI processor chip will help businesses move towards a fraud prevention posture from a fraud detection posture. Source: IBM

Fraud cases on the rise despite new detection solutions

While there are several fraud detection tools available in the industry today, most fraud detection techniques can only detect and catch fraud after an incident occurs. Despite past success, the process can be time-consuming and compute-intensive due to the limitations of today’s technology, particularly when fraud analysis and detection is conducted far away from mission-critical transactions and data.

AI-based fraud detection solutions rely purely on data from past experiences. It looks into the activities, behaviors, and transaction trends for anomalies. In most cases, the AI is able to detect when something is amiss and flag it as a fraud. But again, this only occurs once an anomaly is detected.

Complex fraud detection often cannot be completed in real-time due to latency requirements. Hence, by the time fraud is discovered, the culprit may have already carried out what they intended to do. For example, most credit card companies can only detect fraud after a fraudster has purchased items, or the card is reported stolen.

According to the Federal Trade Commission’s 2020 Consumer Sentinel Network Databook, consumers reported losing more than US$3.3 billion to fraud in 2020, up from US$1.8 billion in 2019. Imposter scams remain the most common fraud reported to the agency.

Imposter scams topped the list of losses due to fraud, at nearly US$1.2 billion reported. Online shopping was the second-most common fraud category reported by consumers, elevated by a surge of reports in the early days of the COVID-19 pandemic, accounted for US$ 246 million in reported losses.

AI chip moves fraud detection to fraud prevention

The AI processor chip is expected to help businesses move towards a fraud prevention posture from a fraud detection posture. Instead of focusing on catching fraud cases, businesses should look to prevent fraud at scale, without impacting service level agreements before the transaction is completed.

As the new chip features an innovative centralized design, clients can leverage the full power of the AI processor for AI-specific workloads. Besides fraud detection and prevention, the chip is ideal for other financial services workloads as well like loan processing, clearing, and settlement of trades, anti-money laundering, and risk analysis.

The chip contains 8 processor cores with a deep superscalar out-of-order instruction pipeline, running with more than 5GHz clock frequency and optimized for the demands of heterogeneous enterprise-class workloads. The completely redesigned cache and chip-interconnection infrastructure provides 32MB cache per core and can scale to 32 Telum chips. The dual-chip module design contains 22 billion transistors and 19 miles of wire on 17 metal layers.

With Telum, the accelerator in close proximity to mission-critical data and applications means that enterprises can conduct high-volume inferencing for real-time sensitive transactions without invoking off-platform AI solutions, which may impact performance. Customers can also build and train AI models off-platform, deploy and infer on a Telum-enabled IBM system for analysis.

With these innovations, clients will be positioned to enhance existing rules-based fraud detection or use machine learning, accelerate credit approval processes, improve customer service and profitability, identify which trades or transactions may fail, and propose solutions to create a more efficient settlement process.

Fraudsters may just find themselves having to do a lot more to wreak havoc on consumers, once the AI processor chip becomes wider accepted in more organizations. The chip may just be the answer the industry needs to not only keep their customers protected, but also ensure their other workloads and financial services can be executed at a faster pace.