Nvidia Accelerates AI Chip Testing Process with Menlo Micro.

Nvidia and Menlo Micro announced on Wednesday that they have achieved a significant speedup in the testing of artificial intelligence chips using the startup's technology. This development is part of efforts to alleviate major production bottlenecks. Nvidia, as a central player in the AI era, is trying to address bottlenecks in its process to meet the enormous demand for its chips.

The company is set to release its earnings report before the market closes on Wednesday, with analysts expecting sales to reach $56.9 billion with a growth of 56%. However, as valuations for AI companies are quite high, investors are monitoring for any signs of a potential bubble bursting.

Nvidia has sold millions of AI chips. Each chip is tested on a special circuit board before sale to determine whether it meets design goals such as speed and other functionalities. However, many of the chips used in these testing circuit boards are over a decade old. This situation complicates the testing of AI chips, which have the fastest communication speeds in the industry and consume large amounts of energy.

To overcome this bottleneck, Nvidia has partnered with Menlo Micro, which split from GE in 2016. Menlo Micro received a total of $227.5 million in funding from Corning and venture capital from Tony Fadell, one of the creators of the iPhone. As a result of their work, a series of switching chips that enhance the performance of test boards has been developed.

Menlo Micro is speeding up the testing processes of circuit boards using metal switches manufactured at the microchip scale. In a research paper published on Wednesday, engineers from the two companies stated that the tests for Nvidia’s graphics processing units (GPUs) could be accelerated by between 30% and 90%.

Menlo Micro's CEO Russ Garcia refrained from disclosing how much business the startup does with Nvidia, but noted that other major chip manufacturers have also started using these switching chips for their test boards. In an interview, Garcia said, "As a result, if you do not validate GPUs before taking the data to a centralized data center, you will encounter errors and other issues. This is the only way to validate these chips quickly."