Diablo boss on chipsets, ULLtraDIMM and the Netlist fracas
Badalone won't back down
Opinion/Interview The Register does not endorse any particular viewpoint contained in this interview. We will update readers on the court's decisions as the case unfolds.
The Netlist/Diablo lawsuit is intriguing if for no other reason than the technical depth of the topic. Diablo's Memory Channel Storage is certainly an interesting product. The Register sat down with Diablo CEO Riccardo Badalone in hopes of working out the details.
The Register: Can you give us some background on Diablo's interactions with Netlist? What's the history that led up to the lawsuit?
Badalone: Our first product was something called an advanced memory buffer, which is literally what it says. It was a memory buffer that isolated the processor and the memory and allows you to expand the memory capacity of the server. We were doing that before we ever met Netlist as a company.
There were at least a dozen companies that started those chips and never were able to finish them. They were nasty, nasty mixed signal, analogue devices.
But we [Diablo] did it. We hit the power, we hit the speed and we got it qualified. With all the other vendors that were already in the market that was actually a very difficult thing to do, but we finished what we started.
So, as a company we had a lot of scars and learning in that space and that's the reason [Netlist] approached us – because we were a reputable silicon company in the memory expansion space.
The Register: So Netlist hired you to develop technology for them?
Badalone: Netlist was started off as kind of a contract manufacturer; none of the management team are really engineering people. In the end there's a difference in technical depth… a huge disparity between the two organisations.
Some of [Netlist's] assertions as a company are "we brought load reduction to Diablo and showed them how it worked, and we had to educate them, etc." Obviously that's not true.
When Netlist started developing HyperCloud, we were one of the suppliers they engaged – they [had] engaged several. We put our technical weight behind that and made it happen in terms of the silicon.
I still think the design that we did for them broke a lot of ground, but it was purely load reduction on the memory channel for memory expansion. That didn't involve flash at the time. It was all about trying to expand the memory footprint of the server.
All of this was this was before we [Diablo] had started gravitating as a company towards the flash business.
The Register: So where'd it all go wrong? Clearly the ULLtraDIMM isn't your standard LRDIMM.
Badalone: The part that [the] Netlist [suit] really centres around is … any time they see chips at the bottom of a DIMM, because they're not a technically deep company, they say: "OK, we do chips at the bottom of a DIMM, [so] they must be using our IP."
I mean, people have been buffering at the bottom of a DIMM since probably 1955. I don't know what it is, but the prior art of retiming signals from one PCB to another has existed since the dawn of time. Obviously they don't have any IP or patents on that.
So in the end what happens is: we announce ULLtraDIMM a year ago, they see buffers at the bottom of the DIMM, and [then] they start essentially making claims that they have IP in the ULLtraDIMM.
The Register: At first glance, HyperCloud and MCS look nothing alike, so what's the claim?
Badalone: I'll give you an example: there's a chip called the RD – the register device that's part of the chipset. And I think that this was in the whistleblower letter. The letter claimed that we basically took the Netlist chipset, spit-polished it and [then] used it. The claim is that it is basically the same chipset that we're using to power the ULLtraDIMM.
The Register: Please correct me if I'm wrong here, but if you tried to use that RD chipset to create an UlltraDIMM, wouldn't it basically explode? Don't you have to address flash in a completely different way than you do standard RAM?
Badalone: So let's talk about that. One of Netlist's big things is "so what's the difference between an RD chip and a 'Rush' ASIC (which is the name of the ASIC that we have in the UlltraDIMM)?
Well, first of all, one difference is about 4M gates. The RD has probably the equivalent of about 40K gates, and they were all placed by hand. RD is an analog design, done by hand, in 130nm technology.
The ASIC that we're using has 4M gates plus an embedded ARM processor, a half a meg of SRAM, serial ports, a full DDR3 physical layer ... it is a full system on a chip.
The Register: It would have to be. There's no other way to convert DIMM signalling into something a flash chip can use. You can't just throw a bunch of flash onto a DDR bus and say "go". You basically have to build a Raspberry Pi and put it between the CPU and the flash.
Badalone: Yep, absolutely. And we have a full protocol that has retries and end to end protection.
The Register: And wear levelling and all the rest of it… You're using consumer grade flash on those DIMMs, right? Which means your algorithms are basically black magic to make sure those things don't fall apart.
Badalone: You're absolutely correct, and that involves other ARMs which are off doing that work. That’s where our flash partner comes in.
So look at the HyperCloud DR chipset. Basically their IP is that they invented a MUX to multiplex between two ranks to fool the processor... Now an interesting thing is they claim we used intellectual property that is in the chipset we provided for HyperCloud in the ULLtraDIMM. I'm talking now about those "buffers at the bottom of the DIMM".
There are some portions of the IP involved that we absolutely did use, because the contract clearly states that the implementation IP is strictly and solely the property of Diablo and we can do what we want with it. We invented that IP. We built it from the ground up. It's amazing technology and we reused it of course in the initial, CPU-facing data buffer for the ULLtraDIMM.
If you look at the claim that we put and the contract that's on the internet, which is public, it literally says that Diablo owns all of the implementation IP. The packages, the masks…the whole thing! This is the basis of our claim back is that "Yeah, of course we're using it", why wouldn't we? It works… and it's ours.
But the stuff that's actually in the claims is IP we did not use. Netlist have this rank multiplication and load reduction technology that they clearly carve out as theirs. We say "OK, it is yours." But, as you clearly know, the ULLtraDIMM has absolutely no DRAM to rank multiply and no loads to reduce. An ULLtraDIMM has only one load!
And so everything that is "proprietary to what Netlist have"... we are not using at all. In the end they started off on a big tirade around HyperCloud and if you notice they're having to completely back off of that … and now they're focusing on NVDIMM.
The Register: NVDIMM?
Badalone: An NVDIMM is a DRAM module that is essentially [a] battery backed device. The reality is we are not building this type of NVDIMM.
The Register: I have to admit that I'm struggling to understand how a flash DIMM that should be committing writes to flash as they happen is anything like a battery backed standard DIMM in anything but the most superficial of ways.
Badalone: They started in discovery, which is not a secret, and they're finding nothing. The best they can do is try to rattle our cage and try to get us to settle, because there's no way they're going to pin any IP or trade secrets on us here.
They basically threw – excuse the expression – a bunch of X, Y, Z against the wall and they're praying that something sticks.
The Register: So why don't you just settle?
Badalone: [T]he reality is that we are not interested in a settlement, because the minute that we indicate that we're even remotely interested in a settlement, it will indicate that we've done something wrong.
Despite the damage this process has done to our reputation, we're still managing to add customers and users. We will see this through in court. We will collect every legal dollar that we've spent and we will collect damages for everything that has been said that is untrue, to our customers, by the Netlist management and the Netlist legal teams.
I'm pretty confident in our position. This is why we said in our press release basically "now we know". We have them up against a wall. We've been very careful not to fight this battle in the press… Last year we were very quiet, making sure we build our case, making sure everything is airtight. Now, I think it's going to turn around and it's going to be a completely different year for them.
The unfortunate thing here is that we've had to invest quite a bit of money in defending ourselves… but my plan is to actually get that all back. And we think we can get it back. So that's how we're going to do this.
In the end, the time-sink of all of this is why most people settle. But, you know, to settle you have to get the other side to the point that they really see that they've lost. Ignorance is bliss: sometimes people don't know when they've lost, so... it has to get worse before it gets better.
The Register sought comment from Netlist, which provided the following canned statement: “Netlist looks forward to presenting its case to the court on December 2nd.” ®
Sponsored: Becoming a Pragmatic Security Leader