I'm talking about these guys: https://www.efinixinc.com Their Trion program seems interesting: - it stretches from area that is occupied by Lattice's MachXO3 on the low end and ECP5 on hight end - no onboard FLASH.Just OTP on few small models and nothing on high end - universal tile that can do routing as well as LUT/MEM/logic - 5 bit BLOCK RAM instead of traditional 9-bit - additional simpplifications on the process end claim 4x chip price reduction ( only 7 metalisation layers instead of 14 etc) Trion on first impression looks nice, but: - a bit slower than ECP5 - based on digikey prices, not cheaper than EPC5, or even pricier at some points has anyone used them and has some data to share on the matter ?
Efinix and their new Trion FPGAs -
Started by ●November 29, 2019
Reply by ●November 29, 20192019-11-29
On Friday, November 29, 2019 at 2:51:51 AM UTC-5, Brane2 wrote:> I'm talking about these guys: > > https://www.efinixinc.com > > Their Trion program seems interesting: > > - it stretches from area that is occupied by Lattice's MachXO3 on the low end and ECP5 on hight end > - no onboard FLASH.Just OTP on few small models and nothing on high end > - universal tile that can do routing as well as LUT/MEM/logic > - 5 bit BLOCK RAM instead of traditional 9-bit > - additional simpplifications on the process end claim 4x chip price reduction ( only 7 metalisation layers instead of 14 etc) > > Trion on first impression looks nice, but: > > - a bit slower than ECP5 > - based on digikey prices, not cheaper than EPC5, or even pricier at some points > > has anyone used them and has some data to share on the matter ?They do seem to be a bit of "me too" in the FPGA arena. Not only are their parts about the same as others already in the field and about the same pricing, they are using all the same BGA type of packaging and the same sort of hard IP. I see the smallest part in the family doesn't have any LVDS I/O. The main complaint I have with them is the near total use of BGA packages. My needs are typically for something like a QFN88 or a QFP100. The QFP144 won't even fit on my board, lol. The T8F81C2 in the BGA81 package is certainly affordable at $5.50. I just wish it had something like a CM0/CM3 processor on the die. I'm going to keep watching Gowin and AGM. -- Rick C. - Get 1,000 miles of free Supercharging - Tesla referral code - https://ts.la/richard11209
Reply by ●November 29, 20192019-11-29
On 11/29/19 10:06 AM, Rick C wrote:> On Friday, November 29, 2019 at 2:51:51 AM UTC-5, Brane2 wrote: >> I'm talking about these guys: >> >> https://www.efinixinc.com >> >> Their Trion program seems interesting: >> >> - it stretches from area that is occupied by Lattice's MachXO3 on the low end and ECP5 on hight end >> - no onboard FLASH.Just OTP on few small models and nothing on high end >> - universal tile that can do routing as well as LUT/MEM/logic >> - 5 bit BLOCK RAM instead of traditional 9-bit >> - additional simpplifications on the process end claim 4x chip price reduction ( only 7 metalisation layers instead of 14 etc) >> >> Trion on first impression looks nice, but: >> >> - a bit slower than ECP5 >> - based on digikey prices, not cheaper than EPC5, or even pricier at some points >> >> has anyone used them and has some data to share on the matter ? > > They do seem to be a bit of "me too" in the FPGA arena. Not only are their parts about the same as others already in the field and about the same pricing, they are using all the same BGA type of packaging and the same sort of hard IP. I see the smallest part in the family doesn't have any LVDS I/O. > > The main complaint I have with them is the near total use of BGA packages. My needs are typically for something like a QFN88 or a QFP100. The QFP144 won't even fit on my board, lol. > > The T8F81C2 in the BGA81 package is certainly affordable at $5.50. I just wish it had something like a CM0/CM3 processor on the die. I'm going to keep watching Gowin and AGM. >I think one reason for the predominance of BGA packages is that they are a lot smaller than non-BGA packages, even with a lot less pins. Yes, there are board cost/assembly issues with BGAs, but my guess is that there isn't enough demand at the very small end to make it worth the tooling cost to make them.
Reply by ●November 29, 20192019-11-29
On Friday, November 29, 2019 at 12:49:13 PM UTC-5, Richard Damon wrote:> On 11/29/19 10:06 AM, Rick C wrote: > > On Friday, November 29, 2019 at 2:51:51 AM UTC-5, Brane2 wrote: > >> I'm talking about these guys: > >> > >> https://www.efinixinc.com > >> > >> Their Trion program seems interesting: > >> > >> - it stretches from area that is occupied by Lattice's MachXO3 on the low end and ECP5 on hight end > >> - no onboard FLASH.Just OTP on few small models and nothing on high end > >> - universal tile that can do routing as well as LUT/MEM/logic > >> - 5 bit BLOCK RAM instead of traditional 9-bit > >> - additional simpplifications on the process end claim 4x chip price reduction ( only 7 metalisation layers instead of 14 etc) > >> > >> Trion on first impression looks nice, but: > >> > >> - a bit slower than ECP5 > >> - based on digikey prices, not cheaper than EPC5, or even pricier at some points > >> > >> has anyone used them and has some data to share on the matter ? > > > > They do seem to be a bit of "me too" in the FPGA arena. Not only are their parts about the same as others already in the field and about the same pricing, they are using all the same BGA type of packaging and the same sort of hard IP. I see the smallest part in the family doesn't have any LVDS I/O. > > > > The main complaint I have with them is the near total use of BGA packages. My needs are typically for something like a QFN88 or a QFP100. The QFP144 won't even fit on my board, lol. > > > > The T8F81C2 in the BGA81 package is certainly affordable at $5.50. I just wish it had something like a CM0/CM3 processor on the die. I'm going to keep watching Gowin and AGM. > > > > I think one reason for the predominance of BGA packages is that they are > a lot smaller than non-BGA packages, even with a lot less pins. Yes, > there are board cost/assembly issues with BGAs, but my guess is that > there isn't enough demand at the very small end to make it worth the > tooling cost to make them.Yes, I'm sure of that, but there are companies selling FPGAs in 100QFP. Now there are a couple of start ups using that package as well as the 88QFN. If you aren't building cell phones, these are alternatives providing a lower system cost not to mention ease of inspection and rework. In particular AGM has 6 kLUT and a 250 MHz CM3 in a 100QFP (if you can buy it) and Gowin has a 4 kLUT part in both a 100QFP and the 88QFN for under $3.40 (a part with a CM3 processor will be out in Q1). -- Rick C. + Get 1,000 miles of free Supercharging + Tesla referral code - https://ts.la/richard11209
Reply by ●December 3, 20192019-12-03
On Friday, November 29, 2019 at 2:51:51 AM UTC-5, Brane2 wrote:> I'm talking about these guys: > > https://www.efinixinc.com > > Their Trion program seems interesting: > > - it stretches from area that is occupied by Lattice's MachXO3 on the low end and ECP5 on hight end > - no onboard FLASH.Just OTP on few small models and nothing on high end > - universal tile that can do routing as well as LUT/MEM/logic > - 5 bit BLOCK RAM instead of traditional 9-bit > - additional simpplifications on the process end claim 4x chip price reduction ( only 7 metalisation layers instead of 14 etc) > > Trion on first impression looks nice, but: > > - a bit slower than ECP5 > - based on digikey prices, not cheaper than EPC5, or even pricier at some points > > has anyone used them and has some data to share on the matter ?I dug into the parts a bit more and if I bite the bullet with the package, the T8 looks pretty good. But it has some limitations. There is 12kB of block RAM available in widths that include x5 and multiples. A bit odd, but potentially useful. So the RAM could be 12 kW of 10 bits. As instruction memory that can be useful. I just wish it had a bit more memory. In the 81 pin BGA package it has only 1 of the five PLLs available and that is "simple" whatever that means. That package only has 55 GPIOs. Many of the I/O features are not available in the smaller packages. The 144QFP is a monster and isn't of much use to me. Packaging is a PITA. -- Rick C. -- Get 1,000 miles of free Supercharging -- Tesla referral code - https://ts.la/richard11209
Reply by ●December 4, 20192019-12-04
On Tuesday, December 3, 2019 at 3:25:45 PM UTC-5, Rick C wrote:> On Friday, November 29, 2019 at 2:51:51 AM UTC-5, Brane2 wrote: > > I'm talking about these guys: > > > > https://www.efinixinc.com > > > > Their Trion program seems interesting: > > > > - it stretches from area that is occupied by Lattice's MachXO3 on the low end and ECP5 on hight end > > - no onboard FLASH.Just OTP on few small models and nothing on high end > > - universal tile that can do routing as well as LUT/MEM/logic > > - 5 bit BLOCK RAM instead of traditional 9-bit > > - additional simpplifications on the process end claim 4x chip price reduction ( only 7 metalisation layers instead of 14 etc) > > > > Trion on first impression looks nice, but: > > > > - a bit slower than ECP5 > > - based on digikey prices, not cheaper than EPC5, or even pricier at some points > > > > has anyone used them and has some data to share on the matter ? > > I dug into the parts a bit more and if I bite the bullet with the package, the T8 looks pretty good. But it has some limitations. > > There is 12kB of block RAM available in widths that include x5 and multiples. A bit odd, but potentially useful. So the RAM could be 12 kW of 10 bits. As instruction memory that can be useful. I just wish it had a bit more memory. > > In the 81 pin BGA package it has only 1 of the five PLLs available and that is "simple" whatever that means. That package only has 55 GPIOs. Many of the I/O features are not available in the smaller packages. The 144QFP is a monster and isn't of much use to me. > > Packaging is a PITA.I tried accessing the web site to get pricing at their store. After days they have not approved my login so I can visit their store. A sales person sent me an email but has not responded to my reply. Strange. I was hoping to get pricing info on some larger parts. -- Rick C. -+ Get 1,000 miles of free Supercharging -+ Tesla referral code - https://ts.la/richard11209
Reply by ●December 5, 20192019-12-05
On 05/12/2019 03:16, Rick C wrote:> On Tuesday, December 3, 2019 at 3:25:45 PM UTC-5, Rick C wrote: >> On Friday, November 29, 2019 at 2:51:51 AM UTC-5, Brane2 wrote: >>> I'm talking about these guys: >>> >>> https://www.efinixinc.com >>> >>> Their Trion program seems interesting: >>> >>> - it stretches from area that is occupied by Lattice's MachXO3 on the low end and ECP5 on hight end >>> - no onboard FLASH.Just OTP on few small models and nothing on high end >>> - universal tile that can do routing as well as LUT/MEM/logic >>> - 5 bit BLOCK RAM instead of traditional 9-bit >>> - additional simpplifications on the process end claim 4x chip price reduction ( only 7 metalisation layers instead of 14 etc) >>> >>> Trion on first impression looks nice, but: >>> >>> - a bit slower than ECP5 >>> - based on digikey prices, not cheaper than EPC5, or even pricier at some points >>> >>> has anyone used them and has some data to share on the matter ? >> >> I dug into the parts a bit more and if I bite the bullet with the package, the T8 looks pretty good. But it has some limitations. >> >> There is 12kB of block RAM available in widths that include x5 and multiples. A bit odd, but potentially useful. So the RAM could be 12 kW of 10 bits. As instruction memory that can be useful. I just wish it had a bit more memory. >> >> In the 81 pin BGA package it has only 1 of the five PLLs available and that is "simple" whatever that means. That package only has 55 GPIOs. Many of the I/O features are not available in the smaller packages. The 144QFP is a monster and isn't of much use to me. >> >> Packaging is a PITA. > > I tried accessing the web site to get pricing at their store. After days they have not approved my login so I can visit their store. A sales person sent me an email but has not responded to my reply. Strange. I was hoping to get pricing info on some larger parts. >They only approved my login after a fairly stroppy email. I think that currently Efinix fail all tests for being ready to do business. I can't understand the package thing - I thought that Microchip had proved that getting design-ins was key to getting business and making it easy to get started was the key to getting design-ins. Sub 1mm pitch BGA packages add huge amounts of cost right at the very front end of a project - often the bit when you're using the tea and coffee budget to try and get the bosses interested. But 0.5mm pitch TQFP and QFN can be hand soldered. I'm amazed that the aspiring newcomers to the FPGA market aren't more actively filling that gap. MK
Reply by ●December 5, 20192019-12-05
On Thursday, December 5, 2019 at 3:47:35 AM UTC-5, Michael Kellett wrote:> On 05/12/2019 03:16, Rick C wrote: > > On Tuesday, December 3, 2019 at 3:25:45 PM UTC-5, Rick C wrote: > >> On Friday, November 29, 2019 at 2:51:51 AM UTC-5, Brane2 wrote: > >>> I'm talking about these guys: > >>> > >>> https://www.efinixinc.com > >>> > >>> Their Trion program seems interesting: > >>> > >>> - it stretches from area that is occupied by Lattice's MachXO3 on the low end and ECP5 on hight end > >>> - no onboard FLASH.Just OTP on few small models and nothing on high end > >>> - universal tile that can do routing as well as LUT/MEM/logic > >>> - 5 bit BLOCK RAM instead of traditional 9-bit > >>> - additional simpplifications on the process end claim 4x chip price reduction ( only 7 metalisation layers instead of 14 etc) > >>> > >>> Trion on first impression looks nice, but: > >>> > >>> - a bit slower than ECP5 > >>> - based on digikey prices, not cheaper than EPC5, or even pricier at some points > >>> > >>> has anyone used them and has some data to share on the matter ? > >> > >> I dug into the parts a bit more and if I bite the bullet with the package, the T8 looks pretty good. But it has some limitations. > >> > >> There is 12kB of block RAM available in widths that include x5 and multiples. A bit odd, but potentially useful. So the RAM could be 12 kW of 10 bits. As instruction memory that can be useful. I just wish it had a bit more memory. > >> > >> In the 81 pin BGA package it has only 1 of the five PLLs available and that is "simple" whatever that means. That package only has 55 GPIOs. Many of the I/O features are not available in the smaller packages. The 144QFP is a monster and isn't of much use to me. > >> > >> Packaging is a PITA. > > > > I tried accessing the web site to get pricing at their store. After days they have not approved my login so I can visit their store. A sales person sent me an email but has not responded to my reply. Strange. I was hoping to get pricing info on some larger parts. > > > They only approved my login after a fairly stroppy email. I think that > currently Efinix fail all tests for being ready to do business. > > I can't understand the package thing - I thought that Microchip had > proved that getting design-ins was key to getting business and making it > easy to get started was the key to getting design-ins. > > Sub 1mm pitch BGA packages add huge amounts of cost right at the very > front end of a project - often the bit when you're using the tea and > coffee budget to try and get the bosses interested. But 0.5mm pitch TQFP > and QFN can be hand soldered. I'm amazed that the aspiring newcomers to > the FPGA market aren't more actively filling that gap.I hadn't thought of it that way. I suppose some number of projects do start with midnight requisitioned, unsanctioned efforts. Still, as the Xilinx reps pointed out some time ago, each package supported adds cost, not just up front, but continuing costs. So any start up will be cautious with adding new packages. Gowin and AGM both are supporting these easier to work with packages, but it is still unclear how "real" they are. As I've posted I have actually had a phone conference with Gowin representatives, but I get zero response from AGM. I don't recall what packages Anlogic has, but other than a board at Seeed has no presence. I'm glad I'm not alone in preferring the QFN/QFP packages. -- Rick C. +- Get 1,000 miles of free Supercharging +- Tesla referral code - https://ts.la/richard11209
Reply by ●December 11, 20192019-12-11
On 05/12/2019 17:02, Rick C wrote:> On Thursday, December 5, 2019 at 3:47:35 AM UTC-5, Michael Kellett wrote: >> On 05/12/2019 03:16, Rick C wrote: >>> On Tuesday, December 3, 2019 at 3:25:45 PM UTC-5, Rick C wrote: >>>> On Friday, November 29, 2019 at 2:51:51 AM UTC-5, Brane2 wrote: >>>>> I'm talking about these guys: >>>>> >>>>> https://www.efinixinc.com >>>>> >>>>> Their Trion program seems interesting: >>>>> >>>>> - it stretches from area that is occupied by Lattice's MachXO3 on the low end and ECP5 on hight end >>>>> - no onboard FLASH.Just OTP on few small models and nothing on high end >>>>> - universal tile that can do routing as well as LUT/MEM/logic >>>>> - 5 bit BLOCK RAM instead of traditional 9-bit >>>>> - additional simpplifications on the process end claim 4x chip price reduction ( only 7 metalisation layers instead of 14 etc) >>>>> >>>>> Trion on first impression looks nice, but: >>>>> >>>>> - a bit slower than ECP5 >>>>> - based on digikey prices, not cheaper than EPC5, or even pricier at some points >>>>> >>>>> has anyone used them and has some data to share on the matter ? >>>> >>>> I dug into the parts a bit more and if I bite the bullet with the package, the T8 looks pretty good. But it has some limitations. >>>> >>>> There is 12kB of block RAM available in widths that include x5 and multiples. A bit odd, but potentially useful. So the RAM could be 12 kW of 10 bits. As instruction memory that can be useful. I just wish it had a bit more memory. >>>> >>>> In the 81 pin BGA package it has only 1 of the five PLLs available and that is "simple" whatever that means. That package only has 55 GPIOs. Many of the I/O features are not available in the smaller packages. The 144QFP is a monster and isn't of much use to me. >>>> >>>> Packaging is a PITA. >>> >>> I tried accessing the web site to get pricing at their store. After days they have not approved my login so I can visit their store. A sales person sent me an email but has not responded to my reply. Strange. I was hoping to get pricing info on some larger parts. >>> >> They only approved my login after a fairly stroppy email. I think that >> currently Efinix fail all tests for being ready to do business. >> >> I can't understand the package thing - I thought that Microchip had >> proved that getting design-ins was key to getting business and making it >> easy to get started was the key to getting design-ins. >> >> Sub 1mm pitch BGA packages add huge amounts of cost right at the very >> front end of a project - often the bit when you're using the tea and >> coffee budget to try and get the bosses interested. But 0.5mm pitch TQFP >> and QFN can be hand soldered. I'm amazed that the aspiring newcomers to >> the FPGA market aren't more actively filling that gap. > > I hadn't thought of it that way. I suppose some number of projects do start with midnight requisitioned, unsanctioned efforts. Still, as the Xilinx reps pointed out some time ago, each package supported adds cost, not just up front, but continuing costs. So any start up will be cautious with adding new packages. > > Gowin and AGM both are supporting these easier to work with packages, but it is still unclear how "real" they are. As I've posted I have actually had a phone conference with Gowin representatives, but I get zero response from AGM. I don't recall what packages Anlogic has, but other than a board at Seeed has no presence. > > I'm glad I'm not alone in preferring the QFN/QFP packages. >You might almost think Lattice are listening: The Crosslink 17 and 40 will be available in 72 pin QFN, as well as all the BGA packages. You only get 40 IO pins out of the 72. MK
Reply by ●December 12, 20192019-12-12
On Friday, November 29, 2019 at 2:51:51 AM UTC-5, Brane2 wrote:> I'm talking about these guys: >=20 > https://www.efinixinc.com >=20 > Their Trion program seems interesting: >=20 > - it stretches from area that is occupied by Lattice's MachXO3 on the low=end and ECP5 on hight end> - no onboard FLASH.Just OTP on few small models and nothing on high end > - universal tile that can do routing as well as LUT/MEM/logic > - 5 bit BLOCK RAM instead of traditional 9-bit > - additional simpplifications on the process end claim 4x chip price redu=ction ( only 7 metalisation layers instead of 14 etc)>=20 > Trion on first impression looks nice, but: >=20 > - a bit slower than ECP5 > - based on digikey prices, not cheaper than EPC5, or even pricier at some=points>=20 > has anyone used them and has some data to share on the matter ?Looking at their manuals they seem to not provide any simulation capability= . They say they are compatible with iVerilog and don't mention VHDL anywhe= re in their documentation that I can find. I guess they only support Veril= og and you are on your own for the simulation capability. =20 Interesting.=20 --=20 Rick C. ++ Get 1,000 miles of free Supercharging ++ Tesla referral code - https://ts.la/richard11209
