Hello! Welcome to Embedic!
This website uses cookies. By using this site, you consent to the use of cookies. For more information, please take a look at our Privacy Policy.
Home > Embedded Events > Broadcom and Google are rumored to work together to design the fourth-generation TPU

Broadcom and Google are rumored to work together to design the fourth-generation TPU

Date: 07-11-2020 ClickCount: 1239

It is rumored that Broadcom and Google have jointly designed the fourth-generation TPU and manufactured using 7nm process. According to a report from "Barron", Google's fourth-generation TPU (Tensor Processing Units, high-speed custom machine learning chips) may still be designed by Broadcom and manufactured using a 7nm process.

Broadcom

JPMorgan Chase analyst Harlan Sur wrote in a report to customers that, like the previous three generations, Google's latest generation of TPU is still designed in cooperation with Broadcom, and this is also the first 7-nanometer processor produced by Broadcom. Currently, the fourth-generation TPU has been put into production, and the fifth-generation design is also in progress. It is rumored that the fifth-generation TPU will use a more advanced 5nm transistor process.

Sur mentioned that Broadcom's application-specific integrated circuit (ASIC) business revenue is expected to reach 750 million U.S. dollars in 2020, while in 2016, this part of the revenue was only 50 million U.S. dollars.

According to Sur's report, in addition to chip design, Broadcom also provided key IP in the cooperation, and is manufacturing, testing and packaging new chips for Google's new data center. In addition to Google, Broadcom's ASIC chip design business customers also include companies such as Facebook, Microsoft and AT&T.

Google launched its first TPU chip in 2016, which is mainly used for data center computing. Customers can use the TPU-based AI platform through Google Cloud Services. The company’s original intention in designing the TPU was to allow this kind of processor to power huge machine learning tasks, and the fourth generation is about three times faster than the previous generation.

Naveen Kumar of Google’s AI department wrote in a blog: “In 2015, even if it took more than three weeks to train AI models with the most advanced hardware accelerators at the time. And just 5 years later, in the face For the same model, Google’s latest TPU supercomputer can increase the training speed by nearly 5 orders of magnitude."

However, Broadcom and Google have not yet responded to requests for comment.

  • Advantech releases TSN switches, opening a new dawn for IT and OT integration
  • Dry goods! Comparative analysis of common embedded processors

Hot Products

  • C8051F389-B-GQ

    Manufacturer: Silicon Labs

    IC MCU 8BIT 64KB FLASH 32LQFP

    Product Categories: 8bit MCU

    Lifecycle:

    RoHS:

  • TMS320DM6467CZUT

    Manufacturer: Texas Instruments

    IC DIGITAL MEDIA SOC 529FCBGA

    Product Categories: SOC

    Lifecycle:

    RoHS:

  • PIC16LF1527-I/PT

    Manufacturer: Microchip

    IC MCU 8BIT 28KB FLASH 64TQFP

    Product Categories: 8bit MCU

    Lifecycle:

    RoHS:

  • TMS320DM6467TZUTD1

    Manufacturer: Texas Instruments

    IC DIGITAL MEDIA SOC 529FCBGA

    Product Categories: SOC

    Lifecycle:

    RoHS:

Customer Comments

  • Looking forward to your comment

  • Comment

    Verification Code * 

Compare products

Compare Empty