Lattice FPGA

The SC series is brand new so you might not find anyone with a socket on their board just yet. It *looks* like a nice family. 3+ Gbit transsceivers. 90 nm. Many 18kb mems. Oh - no "DSP" units in the SC family if I recall correctly - they put those in the "volume" parts instead: the ECP2 family. I like what I see.

If you have high volume needs and want some hard IP on your board, you can get their MACO blocks wired up as a teeny asic in your FPGA.

Hi Jon,

Generally speaking, the LatticeSC compares well with the S2GX and (when available) the V4FX. Though, the MACO blocks on the SCM parts are advantageous over Xilinx and Altera. They provide hard coded functions like a 1GE/10GE MAC, DDR1&2/QDR memory controller, full SPI4.2 interface. All these functions need to be implemented in the FPGA fabric in V4FX and S2GX (consuming a lot of LUTs). The number of I/O's is comparable, but the I/O speed is well above. So you could think that the LatticeSC is superior of V4FX and S2GX. Of course this is my personal opinion, and I can imagine that people in this forum will have other opinions.

Regards,

Luc

Looks nice, but...

If you could only get them (ST2GX or SC).

At least we have shipped all our backlog on xc4vFX devices.

Aust> maxascent wrote:

wau! does it mean all FX MGT issues are solved !?

that would be good news !!

Antti

Are the MGTs up and running? I'd love to prototype with the FX20.

John, Antti,

I will say that if you place an order now, you will get a real delivery date, and the parts will be shipped (we are not supply constrained for the part numbers).

For details, you really need to contact your FAE.

There are more than one sample part number for the FX20, FX60, and I believe now the FX100, that are available.

The FXs with MGTs are rolling out now, so that train is back on its tracks.

In all honesty, there is a lot of characterization left to do, that was delayed because of the changes, so we still very busy with MGTs here.

I have been lately working with some universities and schools in the XUP program to expedite FX parts for them (as they were the last on the list, so if we are getting some MGTs for them, things must be better!).

Aust> Aust>

Austin Lesea wrote

Hmmm. I think John and Antti should be careful here.

We were stuffed with XC3S200 parts because Xilinx' distribution gave us exactly the line Austin suggests. So we ordered a thousand or so thinking we would get a delivery date, and then go ahead or reschedule or cancel depending on the date. What happened was an immediate delivery, and they would not take the parts back. They will finally be used in RockyLogic's next product but one ;-)

So be very careful if Xilinx' outriders in distribution give you the story: "we cannot give a delivery date until you place an order". We have hard evidence that you could end up warehousing Xilinx parts until your project is ready for volume manufacturing. And woe betide you if your requirements change during development - not that ours ever do!

Not good.

Tim,

If you order them, they will arrive. I think that is how it is supposed to work?

Sometimes sooner, sometimes on time (and the objective is to never be late).

So don't blame us that we delivered an order, please!

If you did not want early delivery, then you have to change the terms, and get the distributor to agree to that.

I didn't say (in my previous post) "place an order," I said to contact your FAE for exact details on the part you want.

Unless, of course, you do want a part, then please do place an order.

Aust> Austin Lesea wrote

Austin Lesea wrote

Yes, I do blame Xilinx. Because the line you give out is that a delivery date cannot be quoted until an order is placed. And if we want to discover the date at which volume will be available, we have to place a large order for delivery ASAP. Are you familiar with Catch-22?

I am puzzled by the tone of your response. What I posted was more than amiable, considering the treatment dished out by your distributors (for whom I know you take no responsibility) and you use a public forum to dump your sarcasm on me. That was inappropriate.

Austin,

Tim is right to be p******

1) sometimes placement of an order depends on the known deliver date 2) if ordered items arrive way too early than you have to pay earlier

so speed delivery (before delivery date) may give quite a bit financial problems if money is being used wise. example you know XC3S100E will be delivered W27 so you plan all your actions for the product that uses it in such timeline that money to buy them out will be 'free' for use when the parts arrive, and that the product other components are also purchased at about the same time and production is targetted as well.

now if these parts arrive W14 and not W27 then,

1) you will have to pay immediatly 2) the parts will still be laying around til W27 because that when you get PCB

this is exactly a story I heard from manufacturer of Xilinx based boards.

Antti

Tim,

I did not intend to be sarcastic. Perhaps it sounded so.

It is true that when we were on allocation, there were no dates possible until you were placed in the queue.

We subsequently had a lot which yielded incredibly well (the beginning of a nice and long hard fought for trend).

In any event, I apologize, I didn't intend to offend or insult. Just explain what happens.

If, as you say, we did not give you a date until you ordered, that is consistent with something being on allocation: only those on order have dates. Those not on order are not even in the queue. At that time, we had no idea what the yields would be, so we could not give a delivery date.

Having been so badly beat up (justifiably so) for delivery issues, when we actually succeeded, I was shocked to see your complaint, that is all.

It is a Catch-22, as you say. Works both ways. We don't know either.

Aust> Austin Lesea wrote

Antti,

Personally, I think the purchasing agent should refine their negociation skills.

If you will not accept an early shipment, that has to be specified in the purchase agreement.

Of course, asking the distributor to stock for you (which is effectively what you are doing) will cost them money (evening out the supply/demand), which will raise the price to you.

"There ain't no such thing as a free lunch." (TANSTAFL)

We have some control over our yields, but it is in everyone's best interests that our yields get better and better. If the yield jumps up (defect denisity jumps down) due to a process improvement and learning because we churn out so many wafers, then we will suddenly have all the parts we need. Our our costs go down, our margins go up, and we have more room for negociating prices with our customers.

Sounds like we can never make everyone happy.

Don't yield, we get roasted.

Yield well, we get flamed.

Oh well.

Austin

sorry, it wasnt me complained, but as I had heard a similar story to the posting here, so I posted what I heard. The story was commented with general remark that things with order and deliveries got worse since Avnet swallowed Memec. I dont know the all background, but I can understand that the impossibility to get leadtimes without orders and orders shipped too early can get people upset. Sure it may have been that some one did not read the very fine print.

Antti

Anyone remember the old days, when 'distributor' actually meant what it says ?

-jg

Gentlemen,

The least you can say, is that this thread is quite of the original request from OP. By the time everyone has said what he wanted to say about delivery issues/availability, we are heading up for Virtex10, Stratix10GX, and I'll retired. I find it quite annoying that this happens for every post. (I'm sorry if I'm picking on somebody)

Regards,

Luc

Luc,

We like to talk about things we know.

Sorry.

I myself was really hoping I'd see postings of folks who have GX or SC parts (proving they are real).

Aust> Gentlemen,

Hi Austin,

we had Lattice vice president visiting here last week. The SC parts are real. We did however not got the SC PCIe card loaner yet, all the boards are out, so we are on waitinglist to get to play with the SC board - we are hoping to test 4 lane PCIe on it.

1 maco can do 4 lane PCIe, this IP core solution comes from Lattice, 8 lane PCIe solution will be offered by northwest

SC25 parts should be available, the promise was sample delivery in 3 weeks max if I recall correctly.

So hum, I can assume that the parts actually are real, if the vice talked about boards that are in evaluation by some clients, then there must be parts on them :), but until today I havent seen and SC or ECP silicon yet.

EC, XP, machXO are all real (I have used EC and XP), as of RAM based I would skip EC/ECP and only use ECP2, for nonvolatile solutions macXO if

8by8 mm 0.5mm BGA is needed for space constraints or XP3/XP3 (cheaper than large machXO).

as of SC more, the SERDES is specified up to 3.4G (but is expected to work up to 3.8G)

low cost ECP2 is specified to work with DDR2 400, and can possible be speficied up to DDR2 530 (that depends if... could be, not guaranteed yet). For low cost FPGA its pretty damn good. I have on board on my desk where Spartan3 was considered, but cancelled because of (possible) problems with DDR2 memory.

hear about my fpga logic fabric measurements indicating that S3 is WAY slower than V4 despite using the same technology. Lattice is not using performance reduction on ECP2 so you get the low cost FPGA that works at fabric speed that are normal for the technology. Eg high performance and low cost.

Of course there are things Lattice does badly also, ECP2 has nonvolatile OTP securtiy key, but SC does not have it ! but it would most useable in SC..

Ah, SC15 is possible the only modern high end PFGA with SERDES that is available in FT256 (or same size) footprint - FX12 that is in same sized package doesnt have serdes.

hum, Lattice SoC design environment is coming too, if someone is wondering :) uses Wishbone for interconnect.

Was it On topic now for the OP?

Antti

Hi Antti,

I didn't mean to offend somebody. This post is pretty well on topic. It looks like you did a comparison of SC with V4, but what about S2GX? Could you share some data?

I did some projects with the ECP family and preformance wise they score quite well (esp. DSP) - maybe a bit light on memory. This is also true for the ECP2 - unfortunately.

Based on the datahseet the SC scores well, but lack of DSP and security (like you mentioned).

Best regards,

Luc

schrieb im Newsbeitrag news: snipped-for-privacy@4ax.com...

not much - frst S2GX samples will go out last week march, with more availability in April.

6Gbit serdes PCIe solution provided by PLDA as primary PCIe IP partner.

Antti