The quest for FE310 chips and the future

(Les Marentette) #1

First time poster.

I’ve used the search tool to look for answers to some of my questions, but I’ve not found what I’m looking for.
I’ve also submitted an email direct to Sifive, but not heard back.

Q1) Every photo I’ve seen of FE310 chips has had ‘ES’ on it.
From the datasheet:
"A fully qualified revision of this part in the same package and with a similar pinout
will be available in production quantities. Please consult with SiFive marketing for
schedule and specification."
a) Did FE310 ever complete qualification ?
b) What does the qual schedule look like?

Q2) Where do we buy FE310 chips ?
I know there was supposed to be a website:
But, I don’t see a link to buy chips, just to buy dev boards.

I’m sitting here reading “The RISC-V Reader” and the FE310 datasheet, side-by-side.
From the datasheet I’ve noticed that there are more GPIO than are exposed on the 48pin packaged chip.
It seems with the 48pin version, we loose access to a QSPI2, UART1, and QSPI1 is reduced to single bit mode, since the extra data pins don’t make it out.

Q3) In order to support all of the datasheet identified I/O, it appears to me that a 64pin chip would be needed.
a) Is there a follow-on, higher pin-count FE310 packaged part available ?
b) Is there a timeline for its release ?

I’m really enjoying the RISC-V Reader book, although I’m finding myself reading chapter 2 over and over to really come to grips with the RV32I integer instruction set. I hope the authors someday read Noam Nisan and Shimon Schocken, The Elements of Computing Systems. Also known as the NAND2Tetris Textbook. I find the manner in which Noam & Shimon depict the meaning of each bit, in each machine language instruction for their virtual RISC processor, to be more descriptive and perhaps more intuitive. Maybe that’s just me.

Q4) Final question, with all this processing power and large off-chip NOR / FRAM code space, the FE310 seems to me, to be short on SRAM. Is there a follow-on chip in the works with more on-chip SRAM, or a plan to make one of the QSPI peripherals serve as an addressable off-chip slow SRAM controller ?

Thank You,

(Bruce Hoult) #2

That is the correct link to buy the bare chips. I guess they’re out of stock at the moment.

Most of your questions are answered by the observation that SiFive is not in the business of large scale manufacturing and selling of chips. Or boards, for that matter.

The FE310 and FU540 and corresponding HiFive1 and HiFive Unleashed boards have been manufactured in engineering sample quantities to validate the designs and provide samples to prospective customers who will license the designs in probably customised form for some consumer product that they will (usually) manufacture in quantity themselves. Some chips and boards have been made available to the interested general public via crowdsupply from time to time but the batches are not large and if one sells out then the lead time to another batch of chips is not terribly quick.

It’s possible that at some point either SiFive or a customer will manufacture some standardised SoC configuration(s) in real commercial volume and make them available from Element14/Mouser/Digikey but no plans have been announced.

The FE310 itself taped out about two years ago and is missing a lot of enhancements to the E31 core that is described in the current version of the manual and are available to customers and in the downloadable FPGA bitstream. For example, standard configurations now include up to 64 KB of SRAM. As another example, current versions allow partitioning the icache into both an icache (with fewer ways) and an SRAM instruction scratchpad for critical code.

If you want a single FE310 in a 64 pin package then I think you might be out of luck. If you want a reasonable volume of them then clearly that’s a smaller thing than taping out a new chip. I don’t know the details but it might be worth talking to the sales guys.

(Les Marentette) #3

Thank you for your responses.

I agree, SiFIve does not appear to be in ‘large scale’ manufacturer, but they leave ‘threads’ to be chased like the qualification status of their chips and very specific instructions to request this information in their datasheet.

I do hope that a ‘large scale’ manufacturer will pick up the RISC-V core, add more peripherals and SRAM, but I also fear they will take away the external NOR \ FRAM memory option at the same time. Mainly because I don’t know a single MCU manufacturer that knows how to make good quality, fast and cost effective NOR memory, usually one or more of the vertices on the design triangle suffers. Until some other ‘unplanned’ manufacturer appears on the scene, the SiFive FE310 is the partner that brought us, so it is the one we need to dance with.

Based on your response, SiFIve is not ‘committed’ to long term manufacturing of RISC-V devices on its own, but is instead ‘entertaining’ manufacturing in small scale until one or more interested ‘large scale’ manufactures adopts RISC-V in their own offerings. If that offering is a publicly available MCU, then I’ve been there before…for 130nm \ 180nm process the ‘magical’ target was ten million dollars per year, per part number, in gross sales before a ‘large scale’ manufacturer would take notice. That was ten years ago…the dollar values should be higher now, given all the market consolidation that has taken place since, which likely negates any savings in the tool set.

If the RISC-V core was adopted for internal use, like a configuration controller buried inside a larger chip, then I’m sure the cost savings of eliminating a ‘license fee’ could be motivation enough for a ‘large scale’ manufacturer, but such an implementation will not normally produce a general purpose RISC-V controller for the public. Though…based on your brief comments, that concern may not be part of SiFIve’s exit strategy from ‘small scale’ manufacturing. Perhaps SiFIve will be ‘satisfied’ with a few ‘large scale’ manufactures adopting the RISC-V core for internal use within their devices and simply exit ‘small scale’ manufacturing altogether.

Which takes me to a bit ‘darker’ place today in my thinking about RISC-V, then where I was yesterday.
If SiFive is not committed to growing its manufacturing and support to the very challenging level of success required to attract ‘large scale’ manufacturers to produce and market standalone controllers…then…is the early adopting designer simply setting herself up to ‘take the arrows’ with no reasonable chance of settling on an available RISC-V core ?

Pure opinion: A ‘large scale’ manufacturer is going to be looking for SiFive to surrender their existing and lucrative market base, which SiFive must grow through sweat and toil (generally referred to as backlog) and then use their own incredibly inept marketing team to ‘increase’ the sales number year-over-year (generally referred to as growth). For the first few years, all the new manufacturer will have is the market share SiFive essentially ‘surrenders’ to them to pay their ridiculous salaries and justify their effort to the board of directors. Existing MCU manufacturers, who actually know how to support an MCU core and have the resources to do so will likely be afraid of cannibalizing their existing business. Oh…this is going to be a nasty problem for SiFIve to solve, how to meaningfully move forward without a sizable established market. It’ll be interesting to see if Disney’s Jungle Book significantly changes any of the variables we are accustomed to in modern day Pocahontas. :smiley:

brushing the darkness away

Is it safe to say that the FE310 die already has bond pads for the 13 missing I/O described in the datasheet and maybe some extra VCC / VSS pins ?
Has anyone ever asked for Known Good Die and a bond map ?
If so, Rochester Electronics can whip up a couple 64pins in ceramic package, without having to annoy sales.

Clarifying my personal position \ project, next time I see FE310 chips, I’ll buy a couple hundred dollars worth for my own use. I’ll hope that they are not a last-time-buy, but I’ll solder them into PGA adapters for reuse all the same, because I’m still a little afraid they might just be. I’m definitely not working on the next break-out project that takes SiFive and RISC-V to new heights and fame.

Thank you,

(Bruce Hoult) #4

This is rapidly leaving areas that a mere software engineer is competent to comment on! However I do have a few observations that should be non-controversial.

  • the things that exist today are not the only things that will ever exist. That’s the mistake Steve Ballmer (and many others) made in laughing at the original iPhone, which was indeed quite limited, while at least the 3GS if not the 4 were already in the planning stages. We are currently in very early days for RISC-V commercial manufacture.

  • the cheap SoCs used in things such as the Raspberry Pi or similar boards are the left overs from mass market products. Two excellent examples (and great boards) I happen to know the details of are the Odroid XU4 (Exynos 5422 from the underselling Galaxy S5) and Odroid C2 (Amlogic S905 still used in many cheap set top boxes).

  • SiFive investor Western Digital has publicly committed to shipping a billion RISC-V cores a year. Several SiFive customers announced products last month. See

  • there is not a (singular) “RISC-V core” to be picked up by a manufacturer. There are probably by now at least a dozen entirely different and independent Open Source RISC-V core designs available for download on github, including “darkriscv” which the author reports he developed “in a magic night of 19 Aug, 2018 between 2am and 8am”. There might well be a similar number of closed source commercial cores at various companies as well.

  • RISC-V is license-free in that – unlike x86 or ARM – anyone is legally allowed to develop and use or sell CPUs implementing the RISC-V ISA without getting permission from anyone or paying anyone. Making something production-quality is still a lot of work though, and most companies will prefer to actually pay a license fee to use cores from companies such as SiFive or its competitors (of which there are quite a few).

  • SiFive’s announced strategy is to make a design space of thousands of variations available, and to manufacture (or license) the exact variation each customer wants, on demand. See

(Les Marentette) #5

True, but most of us are not investors, hedging a bet, so what matters most, is packaged parts that exist to be used right now.

This is most likely going to be RISC-V tucked inside a specialized chip intended for storage, not a general purpose RISC-V chip and likely not going to be parts that a design engineer outside WD would have access to.
Maybe that’s where my interest differs from some people interested in RISC-V; I’m interested in parts that I have access to, for the purpose of designing products using standardized packaged components. I’m not contemplating the design of a SoC or hunting for a digital core for use in an FPGA.

Right, I get that, but other than SiFive’s ‘small scale’ manufacturing devices, I’m not aware of any ‘large scale’ manufacturers who have opted to produce and offer for sale packaged general purpose RISC-V controllers for the broad market.

Which speaks to the suitability and popularity of RISC-V for embedding into an SoC, hidden away from general purpose use. There is a lot of cost associated with supporting and promoting a general purpose microcontroller, like Microchip\Atmel, NXP\Freescale, TI, ST and Renesas do today in the broad market. Hopefully a company in India will find an economic model that works for them to be able to do this. I’m not expecting any of the aforementioned companies to do this.

I’ll go read the official announced strategy…having worked in corporate America for 22 years so far, I’ve never seen a publicly announced ‘strategy’, ‘mission statement’, ‘motto’, that had anything to do with what the company was actually doing, planning or trying to accomplish, simply because we didn’t want competitors to know what we were really up to, but maybe this time will be different. :smiley:

Enjoy what’s left of this splendid long weekend,

(Les Marentette) #6

For anyone else who is looking for FE310 chips.

Here is the reply from Crowd Source:
" Thanks for reaching out. Unfortunately the FE310 Chips are not currently available.
If the situation changes in the future, we’ll make an announcement through the HiFive1 page and/or our newsletter.
You may subscribe to updates using the button underneath the video on the main project page:

Eric Strom
Director of Operations
Crowd Supply "

If I get a response from SiFIve on their intentions to make the FE310 available again, I’ll update this thread with that response.

(Les Marentette) #7

If anyone else needs FE310 chips, there are some available here:

For the next 5 days or so.
After that…it ‘sounds’ like it may be Q1 before we see fresh stock.

(Bruce Hoult) #8

FE310 chips are again in stock at CrowdSupply