For those designing for 5V hardware... XC95*XL are gone. Along with Spartan II and Spartan 3, plus more.
https://docs.xilinx.com/v/u/en-US/XCN23009
Darn.
https://docs.xilinx.com/v/u/en-US/XCN23009
Darn.
-
Updated 2023-07-12: Hello, Guest! Welcome back, and be sure to check out this follow-up post about our outage a week or so ago.
AMD (the Xilinx bit) discontinuing a lot of IC's
- Thread starter Melkhior
- Start date
i doubt they care :-( Hobbyist are a tiny fraction of the market for ICs.
Keeping old chips around may have been worth it for an independent Xilinx before the supply chains disruption from Covid, but for AMD they might be more trouble than they're worth - and now they've gotten the axe. I'm worried more EOL'ing might come soon.
What we need are projects to create 5V-compatible chips using SkyWater open-source 130nm PDK (or the other open-source PDKs that are on their way like Glo-Fo's), to replace what's disappearing. But I'm not sure how achievable that kind of goal is. I don't think Google would fund that like they fund current shuttles, those are targetiing "modern" stuff.
Phipli
Well-known member
I feel that is an error. I'm sure ATMEL grew to appreciate Arduino sales. Making something inaccessible isn't the same as there not been a market
What is frustrating is that many fast 5v-3.3v level shifters have also been EOL'd, so it is hard to even do a lower voltage project in the tens of MHz and level shift.
Nixontheknight
Well-known member
I'm sure it's possible to make a level shifter without relying on the IC onesI feel that is an error. I'm sure ATMEL grew to appreciate Arduino sales. Making something inaccessible isn't the same as there not been a market
What is frustrating is that many fast 5v-3.3v level shifters have also been EOL'd, so it is hard to even do a lower voltage project in the tens of MHz and level shift.
Phipli
Well-known member
That's a lot of MOSFETs and resistors to do it with the Philips bidirectional level shifter circuit, on a PowerPC processor card
Nixontheknight
Well-known member
it can't be done with normal transistors?That's a lot of MOSFETs and resistors to do it with the Philips bidirectional level shifter circuit, on a PowerPC processor card
Phipli
Well-known member
Well, yes, but I tend to use 2N7002s for that circuit?
Snial
Well-known member
No way of knowing until it's tried. There must also be a number of FPGA and CPLD architectures which have expired patents now so that will help. There are some SkyWater PDK tutorials I've seen via Hackaday.
skywater-pdk – Hackaday
There are some obviously useful projects that can be done. My priorities would be things like:
- A general-purpose PDS card interface which provides decoding for flash; an MCU interface (both 3v3 ARM Cortex and Arduino) at the very least.
- Built on (1): Classic Macduino, a simple I/O interface that allows direct access to an Arduino's I/O space, so that it's possible to port Arduino sketches to a Classic Mac as drivers (i.e. port the Arduino runtime BSP to classic Mac). I'd favour doing it that way round, because then the only code that needs changing to support different shields (or bespoke I/O) is on the Mac side rather than an Arduino sketch running on the MCU side, and some corresponding driver software running on the Mac side.
- An encryption Co-processor (e.g. any MCU that supports modern encryption). One of the basic problems we have with getting any retro Mac communicating with modern technology is that it all requires SSL x encryption. Most of our solutions involve a Linux Bridge/Firewall which performs thisn, but this means we then definitely need two computers to do what we would have needed one for. Offloading the encryption itself would enable the classic Mac to be fast enough to support secure comms natively IMHO.
bigmessowires
Well-known member
The discontinued ICs include the XC9572XL used in the Floppy Emu. Fortunately I have a lot of them stockpiled while I work on a redesign using a different IC.
I'm not too surprised that Xilinx is discontinuing these, but I am surprised that there's no alternative 5V-tolerant programmable logic device anywhere. I have looked several times (let me know if I missed something). I would have thought somebody would want to serve that audience, even if it's small, and there's some price where it would be worth their while.
Sure you can use a part with 3.3V I/O and level shifters. But that's more parts, more cost, and more PCB area. Rolling your own level-shifting solution with discrete transistors or diodes isn't practical beyond one or two signals. Further complicating things: you may need one shifter IC for inputs and a different one for outputs. Bidirectional level-shifting I/O is challenging if you don't have an explicit DIR signal to control a buffer.
There exist "automatic" bidirectional shifters, which use various methods to attempt to sense which direction the signals should propagate, but they're not foolproof. I wrote about some of the options and the challenges of using them here: https://www.bigmessowires.com/2023/08/22/a-tale-of-three-bidirectional-level-shifters/
I'm not too surprised that Xilinx is discontinuing these, but I am surprised that there's no alternative 5V-tolerant programmable logic device anywhere. I have looked several times (let me know if I missed something). I would have thought somebody would want to serve that audience, even if it's small, and there's some price where it would be worth their while.
Sure you can use a part with 3.3V I/O and level shifters. But that's more parts, more cost, and more PCB area. Rolling your own level-shifting solution with discrete transistors or diodes isn't practical beyond one or two signals. Further complicating things: you may need one shifter IC for inputs and a different one for outputs. Bidirectional level-shifting I/O is challenging if you don't have an explicit DIR signal to control a buffer.
There exist "automatic" bidirectional shifters, which use various methods to attempt to sense which direction the signals should propagate, but they're not foolproof. I wrote about some of the options and the challenges of using them here: https://www.bigmessowires.com/2023/08/22/a-tale-of-three-bidirectional-level-shifters/
I've had success with the 74CB3T family, they're used in nearly all the *FPGA (NuBus needs stronger driver so uses direction-controlled level shifters, which work well but have much longer delays than the pass-through CB3T, fine for 10 MHz but might become a problem at 33+MHz). They are not true 'shifters' though, the 5V side only goes o 3V3 but that's enough to reach TTL 'high' level in practice.
They are also quite expensive. I think they are the #1 source of cost on all my recent designs :-(
Some people have mentioned alternative chips from Lattice that are 5V-tolerant on the EEV Blog Forum, but the software isn't free :-(
Phipli
Well-known member
That circuit might do you if you only have a few connections @bigmessowires?
I use 2N7002 MOSFETs for that circuit. It is commonly used for I2C.
That is a lot of parts being discontinued at once. I see they are accepting orders thru June 29, 2024 with deliveries thru December 28, 2024.
My guess is there are a lot of AI chips that AMD has announced and they want to start building, but there is little available space at their Fabs so they need to make room. I also think AMD has communicated this intent with their large users over the past 6 months and most have decided to switch to newer designs.
My guess is there are a lot of AI chips that AMD has announced and they want to start building, but there is little available space at their Fabs so they need to make room. I also think AMD has communicated this intent with their large users over the past 6 months and most have decided to switch to newer designs.
No, they use completely different fabs. Anything new from AMD is top-of-the line, so single-digit nanometers (MI300 is using TSMC's 5nm for instance).
XC9500XL are made on a 0.35um, or 350nm, process, and I think Spartan 3 are similar. That's the kind that was used for the Pentium MMX... The issue is more likely that making stuff in those fabs is no longer worth it. Between migrating to a new process and shutting down the product lines, they choose shut down, and it does make sense, sadly. If the products are sufficiently in demand from paying customers, there could be a specialized company picking up the designs for continued production (like Rochester Electronics, who can sitll make e.g. MC68030 and MC68040 to support long-term use such as military products).
Even the 7-series Xilinx FPGA aren't bleeding edge enough to compete for capacity with modern GPUs, they use TSMC's 28nm.
bigmessowires
Well-known member
Maybe it depends on which Lattice IC you're targeting? I use the LCMXO2 series in the Yellowstone disk controller, along with the free Lattice Diamond software. You do need to register the software and get a license and renew it every year or so, but there's no license fee.
Yeah, that's handy bidirectional circuit and I use two of them in the ADB/USB Wombat for ADB clock and data. But for something with a dozen I/Os that need to be shifted, it's a lot of discrete parts. You could reduce the part count by using a transistor array and resistor array, but at some point it's probably better to use a purpose-made level shifting chip. That particular circuit is also limited to relatively low speeds since it relies on pull-up resistors for low-to-high transitions. I'm not sure how fast you could push it, maybe a few MHz?
I was thinking about this too. Hopefully somebody will pick it up, or come out with their own 5V-tolerant programmable logic chip. If memory serves, some of the STM32 and ATSAM microcontrollers still have 5V-tolerant versions, at least on certain pins, so the manufacturers recognize there's still demand for 5V.
Phipli
Well-known member
Absolutely, I raised it in the context of the floppy EMU where there aren't that many connections to the host. I was specifically saying packaged level shifters are a better solution in this thread.
But it isn't that slow because there is minimal capacitance, resistance by itself isn't slow.
I've used them on fairly fast SPI circuits on TFT screens. I was generally under the impression that the MOSFET switching time was the main limitation and those things will switch pretty fast.
I haven't pushed it though. I don't tend to do stuff that fast.
Diamond has a free yearly license for some chips (which is not as good as ISE / Vivado, which have free-for-life licenses for some chips), but it seems the 5-V tolerant ispMACH 4000V/Z chips use another (older?) software package, ispLEVER Classic. That used to have free yearly licenses, but no more, they are listed at nearly $600 yearly. This kind of change is why the 'yearly license' model sucks, even when it's free :-( Vendors can stab you in the back whenever they want.
pizzigri
Well-known member
Wouldn’t count on that, some hoarder website may purchase them all and then trickle resell them at like 50x the cost, like many other semi before.
bigmessowires
Well-known member
I've never tried them, but aren't there some open source FPGA design packages that support the older Lattice chips? ICE Storm, ICE Studio, others?
Maybe Lattice believes that anyone who's smart enough to do FPGA development can afford $600/year on software licenses even at the hobbyist level? I don't know. I don't like it either, but these companies have to make money somehow so it's either profits on hardware, software, or both. I looked at their most recent quarterly earnings report, but it doesn't seem to break down earnings from chip sales separately from earnings from software license fees.
Similar threads
