SAN FRANCISCO, Nov 19 (Reuters) - Nvidia ( NVDA ) and

Menlo Micro on Wednesday said they have used technology from the

startup to dramatically speed up the testing of AI chips, easing

a significant production bottleneck.

The world's most valuable listed company and the central

player in the AI boom has been working to iron out kinks in its

processes as it works to feed seemingly insatiable demand for

its chips.

It reports earnings after market close on Wednesday, with

analysts expecting sales growth of 56% to $56.9 billion, LSEG

data showed. Even so, with valuations of AI companies sky-high,

investors are watching for any signs of a bursting bubble.

Nvidia ( NVDA ) has sold millions of artificial intelligence chips,

each of which has to be tested before sale by placing it on a

specialized circuit board designed to determine whether it meets

design goals such as speed and other functions.

Whereas the AI chips are cutting edge, however, many of the

chips in the circuit boards for testing them are decades old.

That makes testing the AI chips, which consume huge amounts of

power and communicate at some of the fastest speeds in the

industry, a challenge.

To address the bottleneck, Nvidia ( NVDA ) has been working with

Menlo Micro, a startup spun out from GE in 2016 and which has

raised $227.5 million in funding from Corning and the venture

fund of iPhone co-creator Tony Fadell. The result is a set of

switching chips that improve the performance of test boards.

Menlo Micro's chips use switches made out of metal, similar

to a light switch on a wall but fabricated at the scale of

microchips using technology from a field called

micro-electromechanical systems.

In a research paper published on Wednesday, engineers from

the two firms said testing of Nvidia's ( NVDA ) graphics processing units

(GPU) could be sped up by 30% to 90% depending on the kind of

test being performed.

Russ Garcia, Menlo Micro's chief executive, declined to say

how much business the startup is doing with Nvidia ( NVDA ) but said

other major chipmakers are adopting its switching chips for

testing boards as well.

"The bottom line is, if you don't validate the GPUs before

you get into the data center, you're going to have errors and

other issues. This is the only way to validate these things at

speed," Garcia said in an interview.