When he joined Rambus in 1991, the company was an unheralded start-up with an idea for making faster PC memory. Nowadays, it's one of the most prominent and controversial semiconductor companies. Advocates say that RDRAM, the name for memory directly based on Rambus' technology, improves overall PC performance because it can shuttle data to the processor faster than conventional memory. At different times, Intel and Sony have both endorsed the technology.
Critics, on the other hand, say RDRAM is expensive and barely speeds things up. Worse, Rambus seems to spend more time filing patent infringement cases against memory makers than it does in the lab. The Los Altos, Calif.-based company is currently engaged in a host of high-profile law suits against Micron Technology, Hynix Semiconductor and Infineon in Germany, the United States and Italy over whether these companies are violating Rambus' patents. Billions of dollars hang in the balance.
The next six months will be a crucial period for the company. Will Rambus convince PC makers and consumers to adopt RDRAM over other forms of memory like SDRAM or DDR DRAM? Or will the industry decide to go with the flow?
CNET News.com recently caught up with Mooring to get his take on the memory market, his company's sometimes close, sometimes not-so-close relationship with Intel, and the company's litigious involvement with the competition.
How did you join up with Rambus?
I joined Rambus in its first year of operation, when there was a CEO and president--Geoff Tate--and a group of engineers. When I first heard about the technology I was in shock. How can they make a DRAM run at 500MHz? When it was ultimately explained to me, I fell in love with the technology and could see all the things that could be ultimately done with it.
It wasn't actually the PC that struck us, first. It was one of the markets. Strangely enough, it was actually larger systems, where they had more memory in them, where we thought the bandwidth was needed first...servers. It wasn't until later that we realized the servers are driven by whatever the cheapest memory is. That led us to games graphics and the personal computers as the ideal markets later, 1992-ish.
The idea came about in 1989 by (Rambus founders) Dr. (Mike) Farmwald and Dr. Mark Horowitz. They spent quite a bit of time, 18 months or so, developing the basis of the technology and then the company was founded in March 1990. I joined as the marketing and sales VP. I was the first business employee (in March 1991).
How is RDRAM different from SDRAM?
The most common type of memory today, SDRAM, has its roots 25 or 30 years ago in what we call a matrix topology, where address and data come in from different directions. That has challenges because wires are different lengths, and it's hard to make it fast when you have one wire that's 1 inch and one wire that's 70 inches. Rambus' topology is based on parallel traces that run more like a racetrack, so every pin is equal distance. That's what allowed us to do the first (product) at 500MHz and now 800MHz going to over 1GHz--because there's a terminated transmission line with identical loads pretty much per pin with a protocol that is used to send address and type of transactions vs. sending them over a different set of control wires.
How did Rambus land Intel as a customer?
It was the third important step and a very important step in the deployment of the technology. The first was just proving the technology could work. What we ran into in the early 1990s was people just thinking we were crazy. They didn't think the technology would work, and they thought our business model was unusual, awkward, unworkable. The first thing we did was prove the technology worked.
The second big event was the commercial usage of the RDRAM...with the Nintendo 64. Because what really made the RDRAM considered a viable high-volume technology was here you had a $150 device that had two of the RDRAMs in it. (Christmas 1995) was the springboard toward Intel choosing the RDRAM in 1996. We had our program kickoff with Intel in 1996. It was a complete redesign. All told, it was almost a three-year development program.
Why does Rambus cost more? What's the history behind that?
There have been huge strides in the price of RDRAM RIMMS (RAMBUS inline memory module) coming down in recent months. There are two reasons for that. One is costs have come down, and the second is prices have come down. The biggest single ingredient to cost reduction is the learning curve of the RDRAM manufacturers. When they started out, it was new and as they built volume, they learned how to lower cost. The second aspect of cost is that we've done some new things...all told, which has brought the cost of the RDRAM solution down to within 20 percent of the cost of SDRAM. But that's different from price.
The price premium of RDRAM's RIMMs was very substantial a year ago, and that was based purely on market supply and demand. Today the price has come down to within tens of dollars. Let's call it a $30 difference for a 128MB RIMM vs. a 128MB DIMM (dual inline memory module). We believe that is a good price for consumers, a good price for (PC) OEMs, and it makes the RDRAM profitable for the DRAM manufacturers. It's the only profitable DRAM that manufacturers can build today. We think it's worth it and...we think educated consumers will find that slight premium worth it, considering the performance advantage.
What about the RDRAM road map? Where do you expect it to go?
(We have) an evolutionary road map on how to apply today's RDAM to new modules that deliver six times more bandwidth over the next five years in stages. There are two technical aspects of that. First is making the RDRAM device run faster, moving from 800MHz today to 1066MHz to 1200MHz. That's a 50 percent frequency gain. The bigger multiple comes from making the modules wider. Just like SIMMS moved to dims and went from 16 to 32 to 64 bits, the RIMM will move from 16 to 32 to 64 bits. So that's a quadruple of performance just by making the module wider but still using the same RDRAM, still using standard connector technology.
Did you take a look at cost?
Cost is a big factor. It's cheaper to change a piece of fiberglass and a connector than it is to redo the entire DRAM infrastructure. The second piece of that is that today, in an 850 platform--Intel's 850 chipset--the chipset that is currently used with the Pentium 4--you have two modules. If you put all of them on one module, you save a couple of dollars. So one module will be cheaper than four if you look out into the 2002 time frame.
How are relations with Intel?
It's a healthy relationship. We have a huge number of people who work with a huge number of people at Intel. The most I can say is that Rambus and our RDRAM partners are working closely with Intel to adopt this road map at the right time for Intel.
Do you have to wait for Intel to introduce new RDRAM technology?
Intel chipsets are available, and they don't talk about their given feature bumps until there's a given time.
Samsung just announced a 1GHz RDRAM module. When will we see it in systems?
The official validated, supported platforms--you'll have to wait for Intel to talk about what's their official plan of record.
What is Rambus doing to promote a less expensive version of RDRAM?
What we do is develop the interface and specifications and then work with the DRAM industry to implement it. With feedback, we realized we could use less banks per DRAM, which would lower cost and still deliver equivalent performance in a PC. So we went from an architecture of what we called 16 dependent banks to four independent banks. There's no visible difference whatsoever, but it helps the DRAM industry build a lower cost core. So the goal is to make an SDRAM-like costing core.
What's the delta between RDRAM and SDRAM at this point, and what does it mean to the consumer?
There's probably different markets. If you're a game player, or heavily interactive visual type of person, you maybe get 30 percent more performance. For me, that's clearly worth the $30. If you're a typist, and you can only type so fast, maybe you don't care about the performance and you save the $30 and buy a better keyboard. But it's their choice.
When you reach price parity with SDRAM, do you believe that RDRAM is the better technology?
RDRAM is clearly the better alternative.
Have any of the issues raised during the trial affected relationship with customers?
There are very interesting dynamics with the RDRAM suppliers today. There are some visible marketing companies like Samsung, Elpida, Toshiba, Windbond. And then there are companies who are very quiet such as Micron, Infineon, Hynix. You'll notice a direct correlation between who has licensed SDRAM and DDR from us and who has not. That's on the legal front. On the business front, it's much like the situation with Microsoft and Apple. Remember where there was a huge lawsuit but they continued to work on applications because there was a mutual need? So even with the companies where there is legal tension, there is engineering work going on. Some of them are quietly shipping RDRAM because it's a profitable part for them.
Infineon claimed at trial that Rambus lifted ideas for RDRAM at a meeting and incorporated them into its patents. Your interpretation?
The two features in discussion were programmable latency and burst. I'm not a patent attorney, but what I know is that when I was marketing the RDRAM back in 1991, those two features were in the RDRAM, and they weren't in any other memory device. So I think there was some inaccurate press about what was the result of the lawsuit, whether it was them incorporating our technology or vice versa. It was clearly our invention.
The biggest part was that we lost the infringement part of the suit, based on what we think was a very poor (ruling), which gave a very distorted view to the jury about who invented what. So if Rambus didn't invent it, because of this poor (ruling), then somebody else must have done it. The whole thing was rather bizarre.
Analysts have said the lawsuits have caused a lot of bad blood between Rambus and the DRAM makers. How would you characterize that?
There's always been a tension for the last decade between the DRAM vendors who don't want RDRAM to succeed and those who do want it to succeed. And that's just standard business. The leaders like Samsung are making more money, and the followers don't like it. The lawsuits have probably slightly exacerbated that, but it's just an inherent tension with market migrations, winners and losers, leaders and followers.
What are you trying accomplish with the lawsuits?
We're trying to get paid for our inventions. When the RDRAM was introduced in 1992, it was radical, revolutionary technology. Over time, some of our inventions leaked into the other technologies (such as SDRAM and DDR SDRAM). We have a choice. We can do nothing or we can try to get paid. We are a public company. We have to protect the rights of our shareholders. There is a lot of stuff we invented. We just want to get paid.
How do you think non-RDRAM chipsets for Pentium 4 will affect you?
RDRAM volume should keep going up. The thing that will change is that RDRAM will no longer be the only memory that works with Pentium 4, and we believe that's OK. The Pentium 4 has to go to lower price points. There will always be people who care only about price and not about the performance they get. Those who care about the extra 10, 20 or 30 percent performance that RDRAM delivers will pay the $30 or $40 price difference.
Can Rambus get by with just income from RDRAM?
Yes. It has been for years and is today a profitable business.
But doesn't that limit your vision for the company?
There are three pieces to our business today: the RDRAM business, the network links business, and then there's the intellectual property business. The IP business will stay about the same size. Where we're growing the company is in high-speed interconnects. We're a chip-connections company. Because we're called Rambus, people will probably always think about putting interfaces on DRAM, but there are lots of other interconnects that we've announced. Plus, we're investing in some other new technologies. One of them, called Yellowstone, we'll be giving a first showing in September.