On Sunday, October 14, 2018 at 8:16:33 AM UTC+11, Klaus Kragelund wrote:> What amazes me is that we are currently looking into ASIC design, and we cannot even get the package for 3 cents, including test vectors. So how do they do an entire micro for 3cents out of the shop?Bankrupt sale. Presumably it was produced - in volume - for a project that failed, and is now being sold off at a price that maximises the short term return. If they charged more, they woud move the stock as fast. -- Bill Sloman, Sydney
3 cent microcontroller
Started by ●October 13, 2018
Reply by ●October 14, 20182018-10-14
Reply by ●October 14, 20182018-10-14
On Saturday, October 13, 2018 at 5:16:33 PM UTC-4, Klaus Kragelund wrote:> What amazes me is that we are currently looking into ASIC design, and we cannot even get the package for 3 cents, including test vectors. So how do they do an entire micro for 3cents out of the shop?They lose a little on each one they sell and make it up in the volume. ;) Rick C.
Reply by ●October 14, 20182018-10-14
On 10/13/2018 11:40 PM, gnuarm.deletethisbit@gmail.com wrote:> On Saturday, October 13, 2018 at 5:16:33 PM UTC-4, Klaus Kragelund wrote: >> What amazes me is that we are currently looking into ASIC design, and we cannot even get the package for 3 cents, including test vectors. So how do they do an entire micro for 3cents out of the shop? > > They lose a little on each one they sell and make it up in the volume. ;) > > Rick C. >"If you go carrying pictures of Chairman Mao, you ain't gonna make it with anyone, anyhow" except bargain-hunting embedded systems engineers I guess! Heh!
Reply by ●October 14, 20182018-10-14
On 10/13/2018 09:31 PM, Klaus Kragelund wrote:> https://lcsc.com/product-detail/Microprocessor-Microcontroller_Bojuxing-Industry-BJ8P509FNB_C83513.html > > 1kW OTP micro, 3 cents >The manufacturers homepage is http://www.bjxmcu.com No update since 2016. Most of the press releases on their homepage mention government incentives, so it seems doubtful that this ever has been a functioning business. Looks like they are defunct.
Reply by ●October 14, 20182018-10-14
On 10/13/2018 11:16 PM, Klaus Kragelund wrote:> What amazes me is that we are currently looking into ASIC design, and we cannot even get the package for 3 cents, including test vectors. So how do they do an entire micro for 3cents out of the shop?Well, the SOT23-6 package they are using is very similar to what is used by discretes. Even at mouser, you can get discrete semiconductors in a SOT23 package for less than 1ct. Negotiated pricing is probably far less. So, yes, you can get a package like this for less than 1ct. https://eu.mouser.com/ProductDetail/ON-Semiconductor/BAV70LT3G Regarding their silicon: Padauk mentions that their lower cost devices are made in a 0.18µm process. Since they are 5V tolerant, I assume that a 5V only CMOS process with minimum metal is used (1PM3). In a minimal configuration this will amount to 15 mask layers, give or take a few. Their core logic is a no brainer and will eat up basically no space, lets assume 0.2mm² to be generous. Pads and I/O may eat up another 0.15 mm². The tricky part is the OTP. They probably need something like 1024x12 bit= 12kbit, assuming that their design is somewhat similar to a PIC12C509. Usually the design for this has to be bought from an IP provider that charges a hefty license fee. Ememory seems to have a suitable product: http://www.ememory.com.tw/html/products_green_neobit.php They claim ~0.05mm² for a 1kx14 OTP Macro. I am not sure whether this already includes supporting circuit, so lets assume 0.1 mm² In total, we end up with 0.2+0.15+0.1=0.45 mm² chip area. Adding seal ring and kerf losses, this would still be less than 1mm² of silicon real estate. Using a low cost foundry it could indeed be possible to source this die for 0.01-0.02 USD. Testing: Lets assume 0.01 USD/pc testing cost. As a bottom line; if you ignore margins and indirect costs (NRE, licensing fees, SGA, distributor fee) you may get close to 0.03USD. The real question is about scaling of the business. These prices only hold true if you sell tens to hundreds of millions of these devices. (A single wafer lots yields >1 million MCUS). So, not sure how this is supposed to work as a business model. I assume that the pricing at LSCS is simply a fire sale. Note that Padauk is a taiwanese company. So they are not likely to receive government aid.
Reply by ●October 14, 20182018-10-14
On Sunday, October 14, 2018 at 12:34:14 PM UTC+2, Tim wrote:> On 10/13/2018 11:16 PM, Klaus Kragelund wrote: > > What amazes me is that we are currently looking into ASIC design, and we cannot even get the package for 3 cents, including test vectors. So how do they do an entire micro for 3cents out of the shop? > > Well, the SOT23-6 package they are using is very similar to what is used > by discretes. > > Even at mouser, you can get discrete semiconductors in a SOT23 package > for less than 1ct. Negotiated pricing is probably far less. So, yes, you > can get a package like this for less than 1ct. > > https://eu.mouser.com/ProductDetail/ON-Semiconductor/BAV70LT3G >Correct> Regarding their silicon: > > Padauk mentions that their lower cost devices are made in a 0.18µm > process. Since they are 5V tolerant, I assume that a 5V only CMOS > process with minimum metal is used (1PM3). In a minimal configuration > this will amount to 15 mask layers, give or take a few. > > Their core logic is a no brainer and will eat up basically no space, > lets assume 0.2mm² to be generous. Pads and I/O may eat up another 0.15 > mm². > > The tricky part is the OTP. They probably need something like 1024x12 > bit= 12kbit, assuming that their design is somewhat similar to a > PIC12C509. Usually the design for this has to be bought from an IP > provider that charges a hefty license fee. > > Ememory seems to have a suitable product: > http://www.ememory.com.tw/html/products_green_neobit.php > > They claim ~0.05mm² for a 1kx14 OTP Macro. I am not sure whether this > already includes supporting circuit, so lets assume 0.1 mm²I looked up the Xfab XC018 process, and 1kROM is only 0,03mm2, so that is not setting the size, it is the ESD cells and the pad grid distance. In this case, the device has only 8 pads, so that does not result in a pad-limited design> > In total, we end up with 0.2+0.15+0.1=0.45 mm² chip area. > Adding seal ring and kerf losses, this would still be less than 1mm² of > silicon real estate. Using a low cost foundry it could indeed be > possible to source this die for 0.01-0.02 USD.Typical cost for wafer, that I know off, is about 0,05 USD per mm2. So 0.025 USD for the die cost alone> > Testing: > Lets assume 0.01 USD/pc testing cost. > > As a bottom line; if you ignore margins and indirect costs (NRE, > licensing fees, SGA, distributor fee) you may get close to 0.03USD. >My guess is that they cannot produce the part for less than 0.03 USD, so nothing left for testing, labour, profit, indirect cost (buildings etc) Cheers Klaus
Reply by ●October 15, 20182018-10-15
On 13/10/18 16:43, Winfield Hill wrote:> Klaus Kragelund wrote... >> >> Dealer site: >> https://lcsc.com/search?q=Microcontroller > > I found a critical Chinese part at LCSC, that's not available > from the usual suppliers, for 30 cents each, and LCSC shipped > 100 pieces by some type of DHL service, for only $10, and they > arrived in four days. Awesome! >If you are happy to go up to $0.30 rather than $0.03, then you can get ARM Cortex-M0 parts from NXP and other more mainstream suppliers.
Reply by ●October 15, 20182018-10-15
On Monday, October 15, 2018 at 9:59:07 AM UTC+2, David Brown wrote:> On 13/10/18 16:43, Winfield Hill wrote: > > Klaus Kragelund wrote... > >> > >> Dealer site: > >> https://lcsc.com/search?q=Microcontroller > > > > I found a critical Chinese part at LCSC, that's not available > > from the usual suppliers, for 30 cents each, and LCSC shipped > > 100 pieces by some type of DHL service, for only $10, and they > > arrived in four days. Awesome! > > > > If you are happy to go up to $0.30 rather than $0.03, then you can get > ARM Cortex-M0 parts from NXP and other more mainstream suppliers.If I was able to go up, I could choose the Silabs EFM8BB1 or ATTiny, they both come in at about 15 US cents, half that of the Cortex M0+ devices
Reply by ●October 15, 20182018-10-15
On 15/10/18 10:00, Klaus Kragelund wrote:> On Monday, October 15, 2018 at 9:59:07 AM UTC+2, David Brown wrote: >> On 13/10/18 16:43, Winfield Hill wrote: >>> Klaus Kragelund wrote... >>>> >>>> Dealer site: https://lcsc.com/search?q=Microcontroller >>> >>> I found a critical Chinese part at LCSC, that's not available >>> from the usual suppliers, for 30 cents each, and LCSC shipped 100 >>> pieces by some type of DHL service, for only $10, and they >>> arrived in four days. Awesome! >>> >> >> If you are happy to go up to $0.30 rather than $0.03, then you can >> get ARM Cortex-M0 parts from NXP and other more mainstream >> suppliers. > > If I was able to go up, I could choose the Silabs EFM8BB1 or ATTiny, > they both come in at about 15 US cents, half that of the Cortex M0+ > devices >There is a vast difference in the quality of the cpu core between these two chips and the M0, if that is important to you. Since you also mentioned ASICs, Atmel (now owned by Microchip) will do you ASICs with AVR cores in them. But I imagine you know that already.
Reply by ●October 15, 20182018-10-15
On Monday, October 15, 2018 at 10:32:00 AM UTC+2, David Brown wrote:> On 15/10/18 10:00, Klaus Kragelund wrote: > > On Monday, October 15, 2018 at 9:59:07 AM UTC+2, David Brown wrote: > >> On 13/10/18 16:43, Winfield Hill wrote: > >>> Klaus Kragelund wrote... > >>>> > >>>> Dealer site: https://lcsc.com/search?q=Microcontroller > >>> > >>> I found a critical Chinese part at LCSC, that's not available > >>> from the usual suppliers, for 30 cents each, and LCSC shipped 100 > >>> pieces by some type of DHL service, for only $10, and they > >>> arrived in four days. Awesome! > >>> > >> > >> If you are happy to go up to $0.30 rather than $0.03, then you can > >> get ARM Cortex-M0 parts from NXP and other more mainstream > >> suppliers. > > > > If I was able to go up, I could choose the Silabs EFM8BB1 or ATTiny, > > they both come in at about 15 US cents, half that of the Cortex M0+ > > devices > > > > There is a vast difference in the quality of the cpu core between these > two chips and the M0, if that is important to you. >By quality, do you mean the Field Failure Rate of the die? I do not expect an ARM device with many more transistors to have less failure rate.... Cheers Klaus
