Article note: Man, I'd really like a little obscene-battery-life 8" class machine like this (for writing and probably also as a terminal) that was well executed and cheap enough not to be precious about, but the E-ink never quite makes it, and bespoke small-market devices are always prohibitively expensive.
I've seen it done with hacked-up kindles a couple times and _those_ seem better than this.
Gadget crowdfunding is great for fulfilling niche demands. Most people, for instance, would never buy a $599 Wi-Fi-enabled typewriter. But in 2016, a Detroit-based startup called Astrohaus proved that not only could such a thing exist, but it could be a delightful and strangely satisfying device. Now, Astrohaus is trying to capture the same experience in a smaller and more convenient package — but unfortunately, its hardware can’t back that promise up.
Astrohaus is the creator of the Freewrite, a chunky word processor with a modern mechanical keyboard, wireless syncing options, and an E Ink screen. Today, the company starts shipping the Freewrite Traveler, a lighter laptop-like alternative. The Traveler is temporarily discounted to $429,...
Article note: General semiconductor industry consolidation.
Keeping up with Intel.
Buying stuff to compete in the current AI hype-cycle.
There are lots of reasonable choices for why they went for this.
Maybe it'll be a good thing, AMD is pretty good about (open) tools lately, so maybe we'll see some tool chain improvement.
After a couple of weeks of rumor, as well as a couple of years of hearsay, AMD has gone feet first into a full acquisition of FPGA manufacturer Xilinx. The deal involves an all-stock transaction, leveraging AMD’s sizeable share price in order to enable an equivalent $143 per Xilinx share – current AMD stockholders will still own 74% of the combined company, while Xilinx stockholders will own 26%. The combined $135 billion entity will total 13000 engineers, and expand AMD’s total addressable market to $110 Billion. It is believed that the key reasons for the acquisition lie in Xilinx’s adaptive computing solutions for the data center market.
AMD CEO Dr. Lisa Su
“Our acquisition of Xilinx marks the next leg in our journey to establish AMD as the industry’s high performance computing leader and partner of choice for the largest and most important technology companies in the world. This is truly a compelling combination that will create significant value for all stakeholders, including AMD and Xilinx shareholders who will benefit from the future growth and upside potential of the combined company. The Xilinx team is one of the strongest in the industry and we are thrilled to welcome them to the AMD family. By combining our world-class engineering teams and deep domain expertise, we will create an industry leader with the vision, talent and scale to define the future of high performance computing.”
Xilinx CEO Victor Peng
“We are excited to join the AMD family. Our shared cultures of innovation, excellence and collaboration make this an ideal combination. Together, we will lead the new era of high performance and adaptive computing. Our leading FPGAs, Adaptive SoCs, accelerator and SmartNIC solutions enable innovation from the cloud, to the edge and end devices. We empower our customers to deploy differentiated platforms to market faster, and with optimal efficiency and performance. Joining together with AMD will help accelerate growth in our data center business and enable us to pursue a broader customer base across more markets.”
Details
As part of the acquisition, Victor Peng will join AMD as president responsible for the Xilinx business, and at least two Xilinx directors will join the AMD Board of Directors upon closing.
Part of the enablement of the acquisition is AMD leveraging its market capitalization of ~$100 billion, and a lot of the industry will draw parallels of Intel’s acquisition of FPGA-manufacturer Altera in December 2015 for $16.7 billion. The high-performance FPGA markets, as well as SmartNICs, adaptive SoCs, and other controllable logic, reside naturally in the data center markets more than most other markets. With AMD’s recent growth in the enterprise space with its Zen-based EPYC processor lines, a natural evolution one might conclude would be synergizing high-performance compute with adaptable logic under one roof, which is precisely the conclusion that Intel also came to several years ago. AMD reported last quarter that it had broken above the 10% market share in Enterprise with its EPYC product lines, and today’s earnings call is also expected to see growth. AMD is already reporting revenue up +56% year on year company-wide, with +116% in the Enterprise, Embedded, and Semi-Custom markets.
The press release states that AMD expects to save $300m in synergistic operational efficiencies within 18 months of closing, due to streamlining shared infrastructure. The deal has been unanimously approved by both sets of directors, and is subject to approval of both sets of shareholders. The transaction is expected to close by the end of Calendar Year 2021.
AMD shares are currently down 5% before the market opens. A conference call will be held at 8am ET to discuss AMD’s Third Quarter Financial results and acquisition plans.
Portfolio
AMD's key product lines includes its Zen based processor lines such as Ryzen and EPYC, its Graphics division for Radeon and Radeon Instinct, and its semi-custom and embedded division which has been developing the latest generation of console processors for both Sony and Microsoft.
Xilinx recently entered the market with its Versal Alveo Adaptive SoCs, built as combination programmable logic plus hardened compute logic and specialized co-processors and accelerators. Its FPGA families include Spartan, Zynq, Artix, Kintex, Virtex, and Virtex Ultrascale, used in a wide variety of commercial, embedded, and enterprise markets, including the hardware used to design processors of the future.
Article note: I still have had 0 Bluetooth devices that haven't at some point done something insane, made difficult to diagnose by lack of human interface. I tolerate BT headphones around the house because non-Bluetooth RF headphones are basically a dead market, but that's about it.
Plus, wandering around in public with BT devices in "HEY EVERYONE LOOK AT MY UNIQUE IDENTIFIER" mode all the time seems like inviting abuse by the usual rogues' gallery of public and private interests.
Article note: The general problem that Wayland is a "Do it all in extensions" design with an inadequate core of features that is already accreting and churning with novelty-seeking developer habits before it reaches feature parity with the creaking hulk of X11 is pretty disappointing.
The fact that it successfully sucked all the air out of the X developer ecosystem while doing so is alarming.
Besides the likes of Red Hat, Intel has been the only other major organization in recent time willing to devote resources to areas like X.Org release management, but even while they let go some of their Wayland folks years ago, they seem uninterested in devoting much in the way of the X.Org Server advancements as we approach 2021. With Ubuntu 21.04 also possibly defaulting to Wayland for its GNOME session, the KDE Wayland support getting squared away, and other advancements continuing, X.Org Server 1.21 may very well prove to be an elusive release.
The transition to Wayland is taking far longer than it should, and a lot of important software simply isn’t ready yet. KDE is still hard at work, and my desktop environment of choice – Cinnamon – has zero support in the works for Wayland. Don’t get me wrong – I’m excited for Wayland – but it feels like we’re counting down by continually multiplying by 0.5 – no matter how many times you multiply, you never quite reach zero.
Article note: Ooh. I don't have time right now, but "more than hacking at it" Ghidra skills are on my skills wishlist, there being a decent reference book should make that learning curve easier.
Article note: The profusion of stm32f103 "compatibles" and their variously legitimate origins and markings is a weird situation, I own a bunch of APM32 black pills (that were properly advertised as such) and like them fine, but it does raise questions.
There is some interesting discussion (that I waded into) in the comments.
Whenever a product becomes popular, it’s only a matter of time before other companies start feeling the urge to hitch a ride on this popularity. This phenomenon is the primary reason why so many terrible toys and video games have been produced over the years. Yet it also drives the world of electronics. Hence it should come as no surprise that ST’s highly successful ARM-based series of microcontrollers (MCUs) has seen its share of imitations, clones and outright fakes.
The fakes are probably the most problematic, as those chips pretend to be genuine STM32 parts down to the markings on the IC package, while compatibility with the part they are pretending to be can differ wildly. For the imitations and clones that carry their own markings, things are a bit more fuzzy, as one could reasonably pretend that those companies just so happened to have designed MCUs that purely by coincidence happen to be fully pin- and register compatible with those highly popular competing MCU designs. That would be the sincerest form of flattery.
Let’s take a look at which fakes and imitations are around, and what it means if you end up with one.
Anatomy of a forgery
Good STM32 IC on the left, clone on the right, with extra dimples.
Earlier this year, Keir Fraser posted an informative summary of some fake STM32F103 ICs as found on so-called ‘Blue Pill’ and similar boards on their Github. The forgeries carry the same marks on the packaging as the genuine STM32 parts, but can often be identified by the pattern of dimples on the packaging, or by the quality of the silkscreen.
These forgeries aren’t always fully functional. As noted by Fraser, many of these parts cannot even be programmed properly, or even run code as simple as the universal ‘blinky’ example. It’s possible that these forgeries are in fact defective STM32F103 dies or similar that are being sold via less-than-legal channels.
The STM32FEB. STM32 it is not.
More insidious perhaps are the near-forgeries that at first glance may look like the real part, but are betrayed by the identification on them: ‘STM32FEBKC6’. That’s not a legitimate ST parts code, and that should be the first tip. This is another clone that’s likely to bring you nothing but grief, as even when it does work, it is a cut-down version of the STM32F103 design, with missing features. Finding detailed information on it is hard as well.
Good artists copy
CS32F103. A more honest clone.
This leaves the trickiest of the clones, in the form of the aforementioned CS32F103. This clone essentially works like the real deal, and can run Blinky compiled for the STM32F103 just fine. Some of these MCUs may even be marked as the ST part, making them hard to identify conclusively.
Some of these are manufactured by CKS (中科芯微), a Chinese company who have apparently made a feature-complete version of the STM32F103, to the point where they have fixed some of the errata listed in the ST datasheet. An article over at CNXSoft provides some more details on this MCU.
A major difference one will quickly encounter with this chip is when programming it and getting the message "UNEXPECTED idcode: 0x2ba01477". The reason for this is that the STM32F103 MCU reports the ID 0x1ba01477, confusing the programmer. This can be fixed for example in OpenOCD by using a configuration script that specifies either no CPUTAPID (0), or this ID reported by the CS32 MCU.
Giga clones
Probably one of the more famous STM32 clone makers is GigaDevice with their GD32 MCUs. As noted over at SMD Prutser in an article series, the GD32F103 appears to be a faster, more capable version of the STM32F103. It has a higher maximum clock speed and faster Flash storage, with a decapped unitshowing that they used two dies inside the package. One for the MCU, and one for the Flash storage, allowing for a rather flexible way to change Flash sizes across their product range.
Decapped GD32F103 MCU. The separate Flash die is visible on top.
At first glance the GD32 MCUs look more attractive than the STM32F1 series, with significant increases in clock speed (72 versus 108 MHz) and Flash storage. While the Flash storage on the GD32 should be very slow, being a serial SPI ROM, its use of SRAM on the MCU die to ‘cache’ the Flash storage means that it ends up being much faster than on-die Flash storage, with zero wait states required even at full MCU clock speed.
A disadvantage of more SRAM instead of pure Flash is that it increases power usage, especially in sleep mode. It also causes a (small) boot-up delay when the SPI ROM’s contents are copied into SRAM before the firmware can run. Depending on the application this may be an advantage or disadvantage. This is of course the same approach as the ESP8266 MCU, which also uses an external SPI ROM for firmware.
When it comes to other GD32 devices, however, they seem to be less eager to make direct clones. Their GDF303 MCU kept the same peripherals as the GDF103, even though those of the STM32F3 are arguably better. This also prohibits their use as a drop-in solution for STM32F3xx boards. Depending on one’s opinion of the STM32F1 peripherals, this may also affect one’s decision to use those GD32 MCUs.
They’re everywhere
Genuine CH32F103 MCU on Blue Pill board.
Although I was aware of the aforementioned fakes and clones, I nevertheless came across a new one recently. This involved the purchase of some ‘Blue Pill’ STM32F103 boards from a big German importer and reseller of all kinds of Maker tat. I wasn’t proud of this, but I needed some cheap boards to use for BlackMagic probes, and they had a good deal. In the comments for the Amazon listing some people mentioned they got a genuine boards, while others mentioned that it was a ‘fake’.
In the spirit of morbid curiosity, I got a couple of these boards and was both horrified and pleased to see that I had in fact received Blue Pill boards that did not carry the promised STM32F103C8T6 MCU, but instead one marked CH32F103C8T6. On the bright side it did not claim to be an ‘ST’ part.
Genuine STM32F103 MCU on Blue Pill board.
This CH32F103 MCU is produced by a Chinese company called WCH, with the (Chinese-only) datasheets and reference manual both provided for download. At a cursory glance, both the datasheet and manaul show a chip that’s practically identical to the STM32F103, with identical memory mapping and peripheral registers.
Hooking it up to an ST-Link/V2 dongle and connecting to it with OpenOCD results in the same CPUTAPID error as with the CS32F103 MCU when using the STM32F1xx profile file. After making the same change to the stm32f1xx.cfg file as suggested by others, I was able to flash the ‘Blinky’ example from my Nodate STM32 project onto the board without further issues.
This suggests that at least the basic RCC (reset & clock control), GPIO and SysTick functionality is similar enough for such a basic test to work. Next, I’ll have to explore whether it also handles the USART, DMA, SPI, I2C and I2S functionality the same way as the STM32F103 MCU that I have on a few other boards. If this MCU is anything like the CS32F103 part, the answer is probably ‘yes’.
As for the seller’s response when I contacted them about these Blue Pill boards not featuring the advertised STM32 part, they admitted that they were aware of this and claimed that ‘in two months’ they’d have boards with genuine STM32 parts again. Admittedly that raises a lot more questions than it answers, least of all why they’d knowingly sell boards that do not feature the advertised MCU.
Time to panic?
The eagle-eyed among us may have noticed that virtually all of these clones involve ST’s first-generation Cortex-M MCUs (STM32F1 series). Unless you need to buy Blue Pill boards for commercial projects, this is unlikely to do more than seriously annoy hobbyists and others who like to have a stack of $3 Cortex-M3 boards around for random projects. If one orders MCUs and development boards from reputable sellers such as Digikey and Mouser, it’s also unlikely to be much of a concern.
The Blue Pill and Black Pill boards are also seeing a bit of an overhaul recently with updated versions featuring STM32F4-based MCUs. Although a bit more expensive than the STM32F103-based counterparts, they do bring considerably more resources to the table and the much more pleasant (in my opinion) peripherals of the STM32F4 line. These may just make the market for the STM32F103 and with it these countless clones, counterfeits, and copies dry up.
Until the first batches of counterfeit, cloned and copied STM32F401 and STM32F411 MCUs hit the market, naturally. Because that’s apparently the name of the game.
Article note: This is an interesting piece, and it doesn't go in any of the directions I expected.
I'm not sure what to take from it, but it's not the dumb entrepreneurial faux-libertarianism the title makes me suspicious of; it argues for compulsory education and institutional formation while arguing _hard_ against aristocracy and organizational calcification (and draws the important parallel between them). Neat.
Article note: Uh... those are power _right of way_ in order to put _power lines_ in them. You know, the power lines that supply power to the neighborhoods, and also will start fires if disrupted?
Yes, the aggressive tree cutting is unsightly, but not doing that leads to power outages and ...getting burnt down by power company negligence like California.
In a rare move, the Lexington council is considering a resolution “roundly” condemning Kentucky Utilities for its “extreme and unilateral tree removal practice” in the city’s rights of way and … Click to Continue »
Article note: I harp on this a lot.
Both in the sense of "know your context, your narrow myopic expertise helps no one but yourself unless you situate it," in the sense that "we need to constrain complexity," and especially in the sense that abstractions rarely truly eliminate complexity, they just hide it in leaky ways that can make the whole ever harder to reason about.
