Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/threads/china%E2%80%99s-analogue-ai-chip-could-be-1-000-times-faster-than-nvidia-gpu.23882/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2030770
            [XFI] => 1060170
        )

    [wordpress] => /var/www/html
)

China’s analogue AI chip could be 1,000 times faster than Nvidia GPU

nature paper here https://www.nature.com/articles/s41928-025-01477-0

making big claims for sure. obviously idea/prototype and scalable system are very different things. geopolitical angle aside, this is one of the effort to replace current power dominated AI infrastructure approach. another one is optical computing, making even bigger advantage claims.

thoughts?
 
I thought ReRAM and PCM were different things?
The Nature paper focuses on ReRAM, just like the IBM one.

Our system uses high-performance RRAM chips that are fabricated in a foundry with a one-transistor-one-resistor (1T1R) cell structure in which each cell has eight conductance levels.
 
Well I think it is a good idea. There are several companies designing these hybrid intelligent memory AI accelerators. Not necessarily with ReRAM.

From what I understand the pioneers of ReRAM were HP with the memristor.

PCM is totally different because it used a process hard if not impossible to integrate with regular logic. It used chalcogenides.
 
Back
Top