One of the interesting areas in the world of new parts recently has been at the lower end of the microcontroller market. Not because the devices there have new capabilities or are especially fast, but because they are cheap. There are now quite a few parts from China under 10 cents apiece, but have the Western manufacturers been able to follow suit? Not quite, but Texas Instruments has a new line of ARM Cortex M0+ parts that get under 40 cents in volume in their cheapest form.
That bottom-of-the-range chip is the MSPM0L1105, a single-core 32 MHz part with 32k of Flash and 4k of RAM. It’s got all the usual peripherals you’d expect on a small microcontroller, but the one which made our heads turn was the on-board 1.45-Msps ADC. On a cheap chip, that’s much faster than expected.
So there’s another microcontroller, and it’s not as cheap as some of its competition, so what? Aside from that ADC there are several reasons to be interested, it has TI’s developer support if you’re in that ecosystem, and inevitably it will find its way on to the dev boards and SBCs we use in our community. It remains to be seen how it will fare in terms of the chip shortage though.
Meanwhile, here’s a reminder of that cheaper competition.
Thanks to the several friends who delivered this tip.
75 thoughts on “New Part Day: TI Jumps In To The Cheap MCU Market”
Honest question for the author/editor, not an attack:
Hackaday isn’t really targetting mass-producing vendors, is it? So, the 1k-unit price of an MCU is much less relevant than the single-unit price to most readers, and as Mrs List notices, this just another entry into the low-cost 32bit MCU market (Kinetis KL03, STM32G030, CY8C…, are all MCUs with at least one ADC > 500 kS/s (wonder what you’d do with higher rates with a core this slow and this little RAM), 32 kB of flash, typically > 2kB of RAM, typically higher maximum clock rates), but with low-volume prices typically above these of TI’s competitors.
So, I wonder, considering this *feels* a lot like a TI press release with a HaD editorial touch, and knowing that Hackaday is part of supplyframe, **whose business model includes helping “leading electronics manufacturers and distributors accelerate new product introductions, […] and take advantage of market opportunities”**:
could you clarify whether or not there was compensation from TI or affiliated companies for publishing this?
Got the same feeling about this article ;(
this is an important question. Thanks to you for having asked it. I’m too interested into reading an answer from HaD to it.
Not a HaD reporter but it seems to me the 1K price break is, indirectly, relevant to hobbyists because dev board makers will get it and it should have direct bearing on the price point for hobbyists.
TI dev support for their MSP430 series chips was pretty good, I’d expect it to be similar level for these too.
Outside of that. Yeah, a lot of the news you read on every tech site will be derived from manufacturer press releases, sad fact of life these days.
TI’s dev boards have also been, traditionally, some of the cheapest on the market. I had no idea why people were using Arduinos at $35/ea when the MSP430 launchpads came out at $4.30.
Those were teaser prices, of course, and over time they’ve crept up and Arduinos have come down. But I’m always on the watch for new TI products to come out just to see how aggressive they are with the dev boards. I still use the Tiva C launchpads that were aggressively priced at under $10 when they announced them after they bought Luminary.
The answer is likely because of the ecosystem arduino built up earlier on (libraries, plug and play modules, dumbed down ide), proliferate documentation and better targeting of the education market. If you were a teacher which would you prefer to base a lesson plan around, a board with many forums catered to with full tutorials on just about every peripheral or one that’s cheaper but maybe has only a fraction of the presence online that would pose a difficulty to students learning about micros for the first time?
It’s the same reason for the success of raspberry pi over cheaper options from chinese sbc manufacturers that sell cheap boards but you are pretty much on your own software-wise with poorly translated (or none at all) documentation. You can have the best hardware and the cheapest price, but without easy to pick up documentation and a low barrier to entry software-side it’ll be an uphill battle when it comes to the masses.
“a board with many forums catered to with full tutorials on just about every peripheral”
That’s why Energia existed. It was also a fantastic teaching tool because you can start off by saying “Arduino’s basically just a set of libraries, not the actual hardware – anything you do using the Arduino libraries can be done on Energia exactly the same,” because Energia was just a reskin of the Arduino IDE anyway.
Then Arduino started to port to other devices, and now you just make a board repository for it and everything runs Arduino anyway. But Energia was really helpful for getting people to be more “device agnostic” for microcontrollers, even if you did have to sacrifice a lot of performance.
Ecosystem. Arduino targeted hobbyists with an easy-to-use IDE and language, making no claims to be ready for industrial deployments.
I’ve got my own tales from the trenches back when Stellaris (later: Tiva-C) launchpads where “hot stuff”, but TI’s vendor HAL was knit with an equally hot needle. TI might have lost a lot of goodwill among firmware designers in those days.
That’s because those weren’t actually TI devices. The Stellaris chips were completely designed by Luminary Micro (hence the “lm4” prefix, meaning ‘[L]uminary cortex-[M4]’, and also hence the “spacey” sounding name). The Tiva-Cs were just a minor revision on those (mostly minor bugfixes). I’m pretty sure the MSP432s were also designed by that same group after they integrated more into TI.
@Pat: ah, Stellaris being a Luminary product series makes sense.
Must admit the MSP432 series never made too much “sense” to me, but then again, I never was the target audience. Branding it as “32 bit alternative to MSP430” really put the product in a tight spot, current draw-wise, and they decided to not go with a small M0+ core, but directly go to M4F (with an FPU! who needs an FPU in an ultra low-power MCU?); I think they were like > 150 µA/MIpS in practice (which is, frankly, still fantastic), and then got beaten out the water by STM32L4 for performance low-power, and by basically any smaller-core STM32L* for anything less compute-mighty…
Bad timing, in the end, probably. Tiva-C didn’t have to happen, honestly: the competitive edge over ST, NXP, Infineon and Atmel/microchip was just not that pronounced. This is a bit of wild speculation, but had they went for developing an MSP430-killer right away instead of taking the detour through an STM32F1/2/4-competitor, MSP432 might have become the dominant ULP family. Welp.
Makes one wonder about the longevity of these new families announced here, doesn’t it?
Again – they didn’t develop the Tiva C (some of the early documentation/APIs still had references to Luminary), and I don’t even really think they developed the MSP432. They just bought it, and in the 432 case I think they just grafted on their existing peripherals from the 430 line to an in-progress design from the Luminary folks try to make it easier for people to transition over.
TI’s own devices were the TMS470 types which were *way* more expensive. They didn’t really have low-cost Cortex designs, which is why they seemed “behind the times.” Because they were.
I’m a big fan of the Tiva C line, actually, mainly because those LaunchPads were just so freaking cheap for the longest time and you could just tell people “download Energia and pretend it says Arduino everywhere.”
TI has been making some of the cheapest micros consistently since the 1980’s, at least. That’s one of the things that comes up sometimes: companies like differentiating on features but if it’s necessary to gain entry into a field, TI can outfab anyone. Quantity has its own quality, as they say.
But inside TI, people use Arduinos on the regular even though launchpads are in theory free. In the time it takes to fire up the arduino ide and program a board and have it running doing what it’s supposed to do, CCS is still loading. Sure, Energia is faster, but you don’t change to another toolchain unless the cost of learning its quirks is lower than the benefits, and the arduino ide is pretty seductive if what you need is something to spit out a digital sequence in a hurry.
Energia *literally* is Arduino. It’s a direct fork. Nowadays you can just pull in the cores themselves into the straight Arduino IDE.
I purchased and played with some of the MSP430 launchpads years back. I never got to the stage of designing any significant projects around them as only a few were available in DIP packages (I’ve since become convertible working with SMD chips), no 5V tolerant I/O on at least the models I played with, low I/O pin sink/source current limit compared to say an ATmega328P at least with the MSP430G2553 chip and such.
I like the low pin current. You could drive an 8segment led display directly. Obviously that’s not the correct way to do it.
Eh? I paid $250 each for TI BQ79606 devboards. Had to get 7 of them to simulate a battery stack. Only felt like they contained $20 worth of parts.
Yeah, their industrial ones are way more expensive.
People use Arduinos because the IDE is more straightforward to use than Code Composer Studio and there are many more “ready-made” examples.
For introducing someone to the embedded world, the Arduino ecosystem is more convenient, but at some point, you have to start learning a professional tool.
Again, benefit of Energia (Arduino). CCS actually allowed you to import Energia sketches (or Arduino sketches!) as CCS projects for easy migration.
Exactly what I thought after reading the first paragraph. There are many things that have an astonishingly low PPU when you buy 1k+…
Maybe we just need something like Drop for MCUs.
On the other hand, some web shop could buy a few batches and resell for ~60 cents and it’d still be a super cheap chip for everyone
… There’s a reason why distributors that literally package millions of different “picks” each day have higher pricing for smaller quantities. It simply costs money to store, cut, mail, handle payment, returns, complaints, customer fraud, not to mention that instead of doing anything else with the money, you invest it in a large quantity of semiconductors which might or might not get a lot cheaper very quickly, and which might or might not see lots of demand.
I, for one, wouldn’t know how to make sustainence-level money from selling components 40ct a piece, even if I could get them for free.
Real simple, sell *lots*
Make it up on volume only works if the customers are willing to pay shopping and handling that, on a small order, may be more than the component’s price.
There “are” vendors willing to do that, for the few buyers willing to do that — mostly buyers who are looking for something obscure, or expect to make larger purchases later from the same vendor. Those I’ve dealt with are perfectly happy to sell prototyping quantities, but the cost per part after S&H winds up being at least an order of magnitude higher than the bulk price, and that really isn’t unreasonable.
HaD announces the bulk price because they can’t predict what the onesie-twosie prices will be. If you want them to wait for that, you find out about the chips much, much later. I’d rather have the early numbers and have to guess from them what retail might be than not have anything it wait for it, but your mileage will vary.
I know amateurs that make bulk orders of cheap parts for future projects.
1k isn’t even really that large an order. I mean, it is for a hobbyist/amateur but even for non-large scale production builds you can burn through several hundred parts pretty easy. Microcontrollers are so cheap at this point that it’s not crazy to put dozens of them on a board.
I’ll say that if whatever you’re building is done a thousand times, you might not be an amateur anymore. Sure, that’s a subjective term, but at that point whether your MCU is 30ct cheaper makes less difference then whether you were able to reduce the cost of the rest of your BOM, and maximize your yield.
Pretty professional-level engineering there.
Not saying there’s not a lot of expert engineers out there that do such things as a hobby, but, hm.
24 microcontrollers per board means even if you only build 10 (which ain’t mass production) you’re at a good fraction of 1000. And 24 isn’t even the smallest number I’ve put on.
24 MCUs per board: OK, that’s interesting! What were you designing? (It’s very unusual, I’ll say, to have 24 MCUs on a single board)
Multichannel control stuff. Doesn’t make sense to have a single big MCU monitor 20+ independent channels, that’s a complicated control system. Instead decouple it, throw a microcontroller at each one, and hey look, you’ve basically got an independent software-defined controller for each path such that the software only needs to deal with basically one thing.
Heck, I’ve thrown microcontrollers on boards just to ease routing! SOP/SSOP-type package microcontroller = remap slow control pins from one side to another one just with a bunch of pin-change interrupts. Way, way, cheaper than, say, a PLD or something similar.
MCUs really should be thought of as more general-purpose ICs than they are.
Heh, I am repeatedly on record telling people that there’s a class of ASICs for control jobs, and it’s called “microcontrollers” :-)
yeah, I agree, if you can divide one large controller into many independent ones, that’s an awesome way to reduce complexity, single points of failures and reduce routing to power rails + maybe 3 wires. Then again, 20+ channel controllers? You sir might be a bit too ambitious to be called amateur ;)
Definitely not amateur, but hacky stuff outside of work gets influenced by experience at work. I don’t think it’s crazy to buy a versatile enough microcontroller in large bulk and then just write specific code to have it replace tons of different things. GPIO expanders, display or key controls, interface adapters, etc.
It’s the same thing old projects used to do with heavy 555 usage.
(or, you, lady, might be too ambitious – not sure of the genus of Pat, not a native speaker)
I share this feeling Marcus.
>wonder what you’d do with higher rates with a core this slow and this little RAM
Oversample like heck.
No, you won’t. All you can do is throw the samples away; memory is too sparse to hold them, and the CPU is too slow to consume them and filter them,
According to the datasheet, it has built in hardware support for oversampling. You can let it average 128 samples, reducing the sample rate from 1.45 MSPS to 11 k samples per second.
Look at Section 9.2.4 – Hardware averaging
You have a valid point but I want to throw in a counter argument. I am a professional and my company makes millions of devices, we use Digi-Key only to get an idea of relative cost because it buying power means that we get huge discounts. I have been reading HAD since I was a poor student and stuff I read here influences the products I design. I really appreciate that my relaxing internet reading also has news that is directly applicable to my job.
Agreed, and it’s not illegimate at all, even, if this *was* somehow compensated monetarily or elsewise – but I’ll be honest, it feels a bit sad I literally can’t tell whether it was or not!
Jenny writes good articles, it’s not fair to her reputation if we don’t know whether this is an “infomercial” or not, and I honestly thought this question, posted about 20 minutes after the article, would have been answered with a simple one liner, I would’ve replied “thanks for clarifying!” and that would have been it. I can live very well with both “Nah, the tip line was just hot and this has nothing to do with how supplyframe earns money” and “Thank you for asking; yes, supplyframe has a contract with TI. Of course, we still make sure our articles are well-researched and we stay true to our hacker ethos. Next time we’ll consider mentioning that in the article itself!” — nothing wrong with earning money through texting. There would be a bit wrong with publishing an advertisement text not mentioning an affiliation.
One of the many reasons that I regularly check Hackaday, is because of product announcments. Whether it was the ESP8266, the ESP32, the TI Launchpads, the newest Raspberry Pi or the Pi pico. Hackaday was always one of the first sites to highlight the release of these and many other parts that Hobbyists regularly use and rely on.
I for one am very excited for these parts. I love the low pin counts and the array of analog peripherals that they contain. With the exception of possibly the STM32F3 family, there really hasn’t been a lot of ARM cortex M microcontrollers with an impressive array of analog peripherals. Especially paired with the m0/m0+ families.
The parts are also being released by TI. They have a lot of experience with analog, DSP microcontrollers e.t.c. And unlike STMicro, they know how to design an SDK that’s actually worth using in a commercial product. So yes, this new family is a big deal for hobbyists and for professionals alike, and worth being highlighted by Hackaday for it’s hobbyist potential alone.
In TI’s marketing release they mentioned the 39 cents microcontroller cost in bulk. Hackaday conveyed that. I see nothing wrong with it.
Accusing Hackaday of being “Corporate shills” of TI or any other company is unfair and unfounded. Hackaday had just announced the much cheaper Puya PY32 about a month ago, along with WCH’s even cheaper RISC-V CH32V003 microcontroller. These announcements help hobbyists know what’s out there and make better decisions. I fail to see how this is Hackday “Doing advertising pieces for TI”. If the next 20 articles where all about the new TI parts, I’d be inclined to agree with you. But they aren’t.
So please whine less….. and hack more!
Really not my intention to whine; I just like my honestly excited news to be distinguishable from advertisement. I’m fine with product announcement! I’m also fine with *paid* product announcements, but it’d feel *a lot* cleaner if these come with a little sticker that says “Ad”. That’s all.
It’s mostly because ti doesn’t set quantity 1 prices so the 1ku price is the consistent reliable price to quote. You can take a look at cut tape prices for parts between digikey and mouser for example and they’ll have pretty comparable prices for a reel but I often find the price for qty 1 to be pretty different.
The 1000 unit price is obviously related to the single unit price. More and more, we hobbyists don’t make out own boards, or even have them made and then source and install the components. Most of the manufacturers of board will install the components for you from parts they have in stock. The ultimate prices they can offer the parts to their customers for depend upon the prices they purchase the components for, which will usually be in units of 1000s rather than 10s. Admittedly, the end user will be purchasing at a markup over 1000 unit prices, but they are relevant.
TI additionally has the advantage of quality over many of the rock bottom priced microcontrollers. I previously worked for a major semiconductor manufacturer on a long term project stress testing a device to monitor the degradation. We monitored the degradation of the parts over two years, simulating the devices operating 10 years or more to refine models of the devices performance.
It should be remembered that semiconductor devices dont last dir ever. They have an expected life time and wear out due to various processes like oxidation and electromigration. It’s striking to watch a signal output that started out an almost perfect waveform degrade over long term testing. For example, you might have a pulse that looks lovely trapezoidal to begin with. You start with a trapezoidal shape a little ringing or overshoot at the corners caused by bandwidth limiting of both your instrumentation and the circuit itself. (I actually think of that as Gibbs Phenomena, which is the mathematical equivalent of bandwidth limiting a sign caused by truncating a Fourier series) Pretty much the textbook waveform. Over years of use (both actual and simulated by elevated operating conditions, you watch that degrade, eventually becoming a mis-shapen little bump and then failing altogether. These degradations are a result of both the process and the design. (Which are very intimately linked) I would expect companies like TI to put a lot more effort into understanding and mitigating these issues than some of the companies whose goal is to make a part for a few pennies. More specifically, I would expect TI to be more forthcoming with exactly what SHOULD be expected from their products in terms of longevity and failure rates and take steps to validate that the meet the specifications.
It should be mentioned, it’s perfectly fine to make a part that for example works reliably for several hundred hours but degrades rapidly and has a 50 percent failure rate in the first year. Just so long as your forthcoming about that. Such a part at the right price point would be fine for example, in a electronic ID badge driving a LCD display for a tech convention. Or a low cost childs toy where it only switches on for a few seconds at a time and spends most of its time idle. Or perhaps even a remote control where the device is powered off most of the time and only powers on for a second or two when the remote control buttons are pressed. But it would be a disaster’s if it was used to build the controller for a microwave oven.
Or 4x times more expensive than ch32v003.
Some progress to replace the WCH flasher by some rp2040:
Or by an esp32s2:
If the MSP430 documentation is any indication, I for one am happily paying an arm (see what I did here? see?!) over asking price for TI’s documentation alone.
The fact that people consider the RP2040 a low-cost MCU has always fascinated me – you need to buy, solder, decouple and program external program memory! Not even counting the fact that this complicates design significantly. Sure, a cheap small SPI flash + RP2040 might still be cheaper than an STM32F0, but neither board space nor extra spins if due to added complexity something goes wrong are free.
In the case of the rp2040, adding a SPI flash IC to the design / BOM is a minor inconvenience at best. A bigger concern for commercial applications is the lack of security features i.e. code on the SPI flash can be easily read off the chip. This can be somewhat mitigated via the use of Secure Flash.
The upside of using QSPI is that you can have up to 16MBytes of SPI Flash / program memory! More than you will ever need! Show me an STM32F0 with that much Flash!
At least we can be happy that Hackaday doesn’t make a post about every “new” MCU from every manufacturer. With the rate at which ST is churning out uselessly similar new STM32 models that cannot be even bought, there wouldn’t be space for even a single hack per day.
How are they “jumping in to” a market they’ve always been in? Their low-end MSP430 parts start at $0.27 by the same unrealistic MSRP metric.
^This as well …
So way back in the dark ages, before the era of arm, I needed a microcontroller. After some research, I found mspgcc, got a pile of MSP430f449 chips, and proceeded. Note the order of operations there… the /actual/ requirement was gcc, and then some chip that could execute its output. Unfortunately since then, TI has outsourced their development tool support to some other company, and they did not impress me with their understanding of opensource. When I attempted to update my development tools, because parallel ports to plug the programming widget into were getting hard to find, the chip support software did not impress me.
My final conclusion was that TI really didn’t appreciate opensource, and would not be supporting how I want to work. Which was disappointing, because as g said, their documentation was pretty solid.
I am guessing you are trolling Tommy, because several open source releases of msp430-gcc and e2e posts by the primary developer at this outsourced company tell a different story. For a number of years they had Josef Lawrynowicz via Mitto Systems working on msp430-gcc. He continued to develop patches for llvm as well.
I am not sure why you parallel port issue has anything to do with the compiler. mspdebug works just fine with the FET430, Launchpad and pretty much any other device TI has put out.
Texas Instruments unveils AM62A, AM68A and AM69A Arm Cortex Vision processors and devkits.
A fast ADC is special, in part, not because its a fast ADC. It’s the inevitable presance of a DMA. And once a DMA is there, it is usually active across a host of peripherals.
So it’s not a microcontroller, it’s a flexible I/O controller for another host. Drive a $1 part over SPI, get an ADC and a bunch of serial/gpio… with the little CPU pushed to its limit just supervising DMA and relaying interrupts.
Wow, a lot of negative comments. Well I am excited. Above all because TI has always provided superb documentation on their parts (in dramatic contrast to some of the datasheets on parts from China for example). For a person doing bare metal programming on parts like these, good documentation is everything frankly and even more exciting than a low price.
Thank you very much for the overview. It was quite helpful.
Many of the remarks here I find rather funny.
In the end we all have to buy the parts,
and to have to find out where to buy 1 each is up to the people who want to buy one.
For me the rough comparison with competition was helpful.
Thank you very much again for the overview.
MSP430G2553 was what I was interested in years ago for a soldering exercise project during a Scout STEM Jamboree.
And TI was kind enough to contribute many hundreds for free for our project.
But new products in DIL are rare now.
TI’s other stuff has been great in my opinion. To me what held the MSP430 back was it didn’t come in enough configurations for package size and on board peripherals. And while I found it easy to program with my prior experience, the dev boards weren’t as entry level as Arduino. Some fun dev kits early on, I still have an MSP430 wristwatch kit.
I hope TI puts out a good kit. But I feel like just making an Arduino compatible board and adding their chip’s support to the IDE’s libraries is both the least they can do and the most they can do.
As someone who was bitten hard by TI’s decision to abruptly disavow and deny their last ARM microcontroller (MSP432) I am more than a little gun shy about this.
did they really?? I’ve dropped out the face of the earth for 3 years on this field
Yeah. As I mentioned above, I don’t think the MSP432 was actually really a TI design. Pretty darn sure it was an in-progress design from Luminary that they finished with quick IP from the 430 line (similar to the Tiva C which was a finished Luminary design that they rebranded).
So it’s not super-surprising that when it wasn’t as successful as they hoped, they canned it, because they didn’t really have a lot of TI resources in it.
As a customer I don’t really care who designed it. It’s got TI’s name on it, so they get the blame for its “interesting” architecture and subsequently abandoning it.
If you had cared who designed it, you wouldn’t have gotten excited about it in the first place. :)
@Pat: Ouch! :)
it’s a shame that e2e is not open to everyone anymore. Been out of the embedded field for 3 years and it was quite surprising. Don’t see a way to get back to launchpads, that community was great
mspgcc did just work… via open source channels, as opposed to TI’s download-the-tools links, which were mostly about installing some commercial compiler (that emphasis is why I have little confidence in TI’s open source entheusiasm). I am also describing events which occurred years ago, I certainly hope things have improved.
The parallel port comment is simply why I went looking to update my development environment.
It’s a shame the whole range only has 1 full featured UART (+1 or more simple UARTS). Kinda makes it useless to me as I need multiple IRDA enc/dec capable ports.
Yes, and I run a dual redundant loop based network between processors. I can do it with 2, but sometimes need to break out.
MCU cost means nothing to engineers- only to their employer.
ARM-core MCU’s are complicated and take significant time to learn, there is no standard for the peripherals. Oh joy another 1,000 page MCU datasheet to learn. Then there is the wait for silicon revisions and bug fixes.
Myself and fellow engineers stick to what they’ve already learned because a new MCU just pushes you behind schedule and puts you on the pressure cooker by management to get the project done.
I think relevant are the quality of the toolchain and documentation, as well as age of the part – how many bugs are flushed out and known. I have a bitter taste from the LM3S Tempest FLASH failure debacle and does TI have decent tech now?
You’ll spend $1,000’s on engineering wages on time to learn the part and design it in – verses the cost savings on the cheaper MCU on build quantities. It’s actually not worth it, when you do the math to save $2 on the MCU verses the weeks of head banging.
While an MCU with a (reasonably) fast ADC is nice, I’d prefer a (significantly) fast ADC with some DSP.
There are ARM chips clocked at 150-200MHz. A 160M ADC with some smarts could downsample to 40M or lower, reducing the need for a high speed interface to the next processing stage.
Generally most of these guys have support for built in hardware averaging.
Going above about 10 MSa/s typically requires a completely different ADC architecture, because for 12-bit resolution you need roughly 10x the rate in something like an SAR, so like 100 MHz. Obviously going to 160 MSa/s would push that SAR into the GHz range, which isn’t practical, so 160 MSa/s ADCs are flash ADCs, which take up *way more* die area than an SAR. (Yes, there are 15-18 MSa/s ADCs on chips, I said “about”).
Your options for chips with on-die flash ADCs are basically zero – as far as I know, there’s the LPC4370 from NXP at 80 MSa/s, and that’s basically it.
Let me guess… “Limit 50”
“TI’s developer support” lol
Please be kind and respectful to help make the comments section excellent. (Comment Policy)