Rambus better RAM secret sauce = ASML good news "indirectly, but strongly"
Copilot:
In particular, demand for memory interface chips from Rambus has been voracious . Memory chip revenue has climbed 42% per year over the past five years as Rambus has expanded its portfolio of products in the area. Those results have come largely because of the importance of Rambus' dynamic random access memory (RDRAM) in the hierarchy of memory solutions. Data centers rely on hard disk drives and solid state drives for large volumes of data, but they're much slower to access than the RDRAM and other novel memory solutions that Rambus has embraced.
The Rambus recovery could be for real
You’re reading a passage that mixes old RDRAM history with modern Rambus success, and the key is to separate the two. The “secret sauce” today has nothing to do with 1999-era RDRAM modules. Rambus no longer sells DRAM chips at all. What’s exploding is Rambus’ memory-interface IP and PHYs, not RDRAM devices.
Let’s break it down cleanly.
1. What RDRAM was (1990s–early 2000s)
RDRAM’s original “secret sauce” was:
- High-speed serial interface instead of a wide parallel bus
- Very high clock rates for the era
- Packetized, pipelined transfers
- Low pin count vs. SDR/DDR
- High bandwidth per pin
This is exactly what the Wikipedia entry describes: a serial, high-frequency, narrow-bus DRAM architecture designed to beat SDRAM on bandwidth.
But RDRAM lost the standards war because it was:
- Expensive
- Proprietary
- Hot (needed heat spreaders)
- High latency
- Politically toxic after Intel backed away
So RDRAM is not what’s driving Rambus’ current revenue boom.
2. What Rambus’ “secret sauce” actually is today
The modern Rambus growth engine is:
A. High-speed memory interface IP (PHYs + controllers)
Rambus sells:
- HBM3/3E PHYs
- GDDR6 PHYs
- DDR5/DDR6 PHYs
- PCIe 5/6 SerDes
- Security IP
This is the “memory chip revenue” the article refers to — not RDRAM chips. Rambus’ PHY IP is embedded in:
- AI accelerators
- GPUs
- Networking ASICs
- Automotive SoCs
- Data-center CPUs
This aligns with the fact that Rambus PHY IP is found in HBM stacks, GDDR6 systems, DDR5 controllers, PCIe 5/6 SerDes, and AI accelerators.
B. The real secret sauce: signal integrity at insane speeds
Rambus’ competitive advantage is:
- Ultra-high-speed I/O design
- Clocking, equalization, and training algorithms
- PHYs that can run reliably at 20–30+ Gb/s per pin
- Low-power, low-jitter, low-latency interfaces
This is exactly the bottleneck in AI systems today: feeding GPUs and NPUs with enough bandwidth.
Rambus is basically the Allen-Bradley of memory interfaces — the company you call when you need deterministic, high-reliability, high-speed signaling.
3. Is there an ASML angle? Yes — indirectly, but strongly.
A. HBM drives EUV demand
HBM is the fastest-growing DRAM segment, and it requires:
- More EUV layers
- More advanced DRAM nodes
- More 2.5D/3D packaging
We’ve discussed before that HBM is now consuming ~20% of all DRAM bits and forcing DRAM makers to shut down consumer DRAM lines. That shift increases EUV demand for DRAM fabs.
B. Rambus benefits from HBM adoption
Rambus doesn’t make HBM chips — but it sells the HBM PHYs that connect them to GPUs and AI accelerators.
So the chain looks like this:
AI ? HBM demand ? more EUV layers ? more ASML tools AI ? HBM/GDDR/DDR6 interfaces ? more Rambus PHY licensing
Both companies ride the same megatrend from different layers of the stack.
C. Rambus’ success is a leading indicator of memory bandwidth stress
When Rambus revenue spikes, it signals:
- GPUs are starving for bandwidth
- HBM adoption is accelerating
- DRAM makers must scale advanced nodes
- ASML’s memory-side EUV demand rises
So while Rambus and ASML don’t interact directly, they are tightly coupled through the physics of bandwidth.
4. Why the article mistakenly invokes “RDRAM”
The passage you quoted is likely using “RDRAM” as shorthand for “Rambus memory technology,” but that’s outdated. Modern Rambus revenue comes from:
- HBM PHYs
- GDDR6 PHYs
- DDR5/DDR6 PHYs
- SerDes and security IP
Not from RDRAM chips.
Bottom line for you, Mike
RDRAM’s secret sauce was serial, high-speed, narrow-bus DRAM — but that’s ancient history. Rambus’ modern secret sauce is world-class high-speed PHY design for HBM, GDDR, and DDR6. The ASML angle is indirect but powerful: HBM demand ? more EUV ? more ASML revenue.
If you want, I can map Rambus’ PHY roadmap (HBM3E ? HBM4 ? GDDR7 ? DDR6) to the DRAM node transitions and EUV layer counts to show exactly where the ASML leverage sits.
PS
Better techniques bonanza JUST started.
The bonanza of bonanzas JUST started.
ASML
Village |
