Elroy, yes, there's a very positive mood for HBM - I just listened to Micron's conference call and there's a
lot of fanfare about HBM and its future products like HBM4e towards the end of the decade.
Rambus definitely makes money on their HBM IP, but not as much as people thought - including me.
But, don't get me wrong, it's an important growth driver for Rambus and it's great for them to be in
the middle of all of this major HBM growth cycle driven by Nvidia and the companies who are trying
to catch up to them.
The revenue that Rambus collects from their HBM controllers shows up in their silicon IP business.
From the products alone that business is about $115 million annually - add in the fixed fee patent
portion of that and their annual contribution of silicon IP is about $200 million. The fixed fee
patent portion is really important because both Samsung and Hynix signed up for a decade
about 2 years ago. Micron is due for renewal as well and they should sign up before the
year is out. In aggregate, that business is about $220 million and stable because it's fixed fee.
But it provides a lot of excellent cash flow for Rambus to work with their customers on the
many complex problems that the industry needs for things like HBM growth, GDDR6 and
beyond, memory modules, CXL, etc.
The good thing for Rambus shareholders is that we've been told many times that the silicon IP
business should grow at 10% to 15% annually. So, everything Rambus does with HBM controllers
(as well as GDDR6 controllers, PCIe, CXL, etc) shows up in their silicon IP revenue - here they are:
Interface IP - Rambus
The good thing is that this should grow like clockwork because there's so much business that Rambus
has expertise in for these areas - just check here for HBM3e:
HBM3E: Everything You Need to Know - Rambus
That's not the only area of growth for Rambus - this below is doing quite well too and Rambus expects
to have 40% market share in the DDR5 market for their registered clock drivers and a 20%
share for their companion chips - right now the TAMs for those markets is about $750 million
and $600 million respectively and growing. The companion chips will grow even more when
the AI PC proliferates and the memory modules will need these chips at and above 6,400 MT/s
(PMIC, SPD hub and temperature sensors) for the memory modules.
Memory Interface Chips - Rambus
Also, Rambus is thinking big about the PMIC market as it relates to memory - and the buffer (RCD)
market too - we can see in Rambus HBM patents that buffers will be needed for the communication
between logic and memory.
And lastly there's this - Rambus has already prototyped a sophisticated controller chip - but
this won't be an important growth driver until CXL 3.0 comes out in the 2026 timeframe and
forward. It's really a 2027 plus growth opportunity. But Micron (and many others) think that
CXL can be a big deal to help optimize the data center:
CXL Memory Initiative - Rambus
Rambus' Larry Carr is on the board of directors and he's also the president. I think that's
because Rambus' expertise in serial attached memory goes back decades:
Homepage - Compute Express Link |
