Broadcom has demonstrated that it is probably the world's largest processor. But for what purpose? When you visit a TSMC event, you will always see multi-chip products using the company's Chip-on-Wafer-on-Substrate (CoWoS) packaging technology, with features close to the limits of reticles (858mm^2, 26mm x 33mm). The deck of the let processor was shown. mm) Calculate the chiplet. I can't take pictures of the deck, but there's definitely a processor that will grab your attention. One of these devices is made by Broadcom and was unveiled at the company's recent investor event.
To most observers, Broadcom is a networking and communications giant, but the company also has a significant custom chip design business. For those unfamiliar with this division of Broadcom, Google is one of the company's most prominent customers in terms of contract chip designs.
However, like TSMC, Broadcom also does not announce clients. For those looking to reignite innovation in the short term, Broadcom lists them in a recent press release. What the company does to make a good impression is show investors its vast achievements. As our friend and colleague Patrick Moorehead of Moore Strategic Market Analysis has observed, these are vast indeed.
“Here’s to another fun thing,” Patrick Moorhead wrote in an X post. “The smiling man, Frank Ostosich” [who] Run a custom silicon group from Broadcom. He should be smiling when he announces that he's acquired his third XPU design from a major “consumer AI company.”
There's another fun thing to do. The Laughing Man (Frank Ostojic) runs @Broadcom's Custom His Silicon Group. He should be smiling when he announces that he's acquired his third XPU design from a major “consumer AI company.” On the right is a close-up of his XPU. You will see two compute units… pic.twitter.com/sseCi02B4KMarch 21, 2024
Broadcom officially brands these chips as XPUs to avoid exposing applications. On the other hand, the use of high-bandwidth memory is pretty much indicative of its target use, which could possibly be network switching infused with artificial intelligence or hardcore AI.
“On the right is a close-up of the XPU,” Moorhead added. “In the center he sees two compute units and on the left and right all the HBMs. A lot of compute, HBMs, very fast on-chip connections, and as you can imagine, a complete set of units with the highest performance. External Networking for Custom SoC Performance.”
The development of chiplets of this scale (i.e., close to reticle size) has already been achieved. Getting it to the right level is another dimension of accomplishment, and it looks like Broadcom's foundry partner, probably his TSMC, has accomplished that as well. Now is the time for software to catch up and take advantage of the power of this processor.