Electronics > Projects, Designs, and Technical Stuff
Oscilloscope ASIC Kickstarter instead of Open scope.
ogden:
--- Quote from: excitedbox on January 10, 2020, 03:29:04 am ---The negativity I refer to is the stuff below that I am answering for the 8th time now. The one guy brings up a $900 and in his next post claims no manufacturer could afford to sell a scope for that price.
--- End quote ---
They can't sell "fully unlocked" scope for price of BW-limited one! With same success you can ask all software companies to sell their full products for price of "lite" version. Your argument is not even naiive. It is absurd.
David Hess:
You can accomplish everything you suggest with a reference design using existing parts including a bit of programmable logic and no custom ASICs. Nobody has even done that so I doubt the economics make either viable and an ASIC based solution even less so. Full documentation to an existing DSO with the proper hardware would allow it but that is sure never going to happen.
--- Quote from: unitedatoms on January 10, 2020, 01:34:41 am ---From years ago what I remember about advertised breakthroughs in oscilloscope ASICs is two things: One brand advertised 50000/sec screen updates, may be meaning the good coverage of triggered sweeps ratio to missed events. Another manufacturer claimed trade mark of "Digital phosphor". That made me think that actual challenge in scope ASIC is dual-or-multiport memory with digital adders and faders per "pixel" (bin of Y magnitude). Say for Megasample wide X window you actually need full vertical Y column of "pixels" with all accumulated counts and fading. Perhaps 500x16 bit values. Before any optimization and compression that means 1GByte of multiport SRAM. That is core of ASIC. The rest are small things like interpolators, non-linearity correction by lookup, addressing routes/pipes to feed the SRAM, adders, faders, output interface to present SRAM to external CPU bus.
The multiport SRAM can be turned into external normal DRAM may be, say put large amount of DRAM chips on individual buses for parallelism, as much as pincount allows.
Then ASIC becomes a deserialising data sorter, updater to external DRAM, plus some DMA from the point of view of CPU.
I am just theoritizing about what "universal" can there in such pure ASIC for reusability as pure digital chip. May be such chip or major part of such chip already exists in form of say obscure telecom "Bla ATM arbiter, backbone router, massive crossinterconnect Cisco Bla thing"
--- End quote ---
The ASIC or programmable logic generates a 2D histogram in real or decimated time. Performance is limited by memory bandwidth so embedded memory which can be wide and shallow has an advantage over external memory. High performance desktop CPUs have enough cache bandwidth and instruction parallelism to take advantage of it to perform this operation.
Some compromises in performance would allow a commonly available CPU for embedded use to produce a reasonable display assuming that the digitized data could be delivered to it.
wraper:
--- Quote from: excitedbox on January 10, 2020, 03:29:04 am ---
--- Quote from: wraper on January 10, 2020, 02:37:41 am ---
--- Quote from: excitedbox on January 10, 2020, 12:20:10 am ---2. Manufacturing costs are tiny in comparison. (AMD pays less than $13 for a CPU made on the MOST expensive node while this will be one of the cheapest. A qty discount can only go so far since there are still set costs such as materials and EUV is very slow)
--- End quote ---
Do you really believe that? If that was true, they would sit on piles of money instead of going towards bankrupt before releasing Ryzen. Such price is valid only for their outdated low end CPUs on old node.
--- End quote ---
No, those prices are for their 12nm node which are much more expensive and these asics would be made on much older nodes anyway due to the higher voltages and cost savings. At the smallest 45nm but I am assuming either 65nm or even larger. That would have to be determined.
--- End quote ---
Please don't make such bolt statements without providing anything supporting it, showing source of this data. Googling does not show up anything supporting this claim.
--- Quote ---Up until 2012 Intel was paying Dell up to $1.2 billion a year not to use AMD chips.
--- End quote ---
As if it was enough to cause this. And I knew about it even without you. Also I knew that Intel paid compensation to AMD.
--- Quote ---AMD tried gifting HP 1 million CPUs FOR FREE
--- End quote ---
Now it's simply BS.
EDIT: BTW HP made AMD based desktop PCs and laptops before 2012 when judgement was made and before 2009 as well. have seen plenty of them.
excitedbox:
You can not say what can or can not be afforded without knowing how much this chip costs. We know right now the ADCs etc are expensive. We know FPGA development takes a lot of time, making it expensive. We know Open Source projects reduce costs in 95% of cases. Paying 5-10 engineers 50-60k a year to develop memory controllers on an FPGA is money you can save if you have a chip that has a memory controller on it.
David Hess- Keysight, Tektronix, Rigol, all have their own ASICs. Did you even read the first post? You picked single pieces of information to base your argument on and didn´t look at the whole picture.
You say there wouldn`t be competition but when anyone can buy the same chip, if you don´t compete someone else makes the same product and your share is reduced. Also there are many ways to skin a cat. I might want to save a few bucks and buy a scope with USB 3.0 while someone else wants a larger LCD and another person wants even faster data transfer to chain instruments. Most of the things that add value to a scope are beyond the signal capture. Signal capture is only the first step. Even in the UI there are enough differences that 10 people want 10 different devices. Some people love USB scopes while others would never buy one.
Someone mentioned a ~$1300 scope with 350MHz. Hantek has a $600-700 scope with 300MHz. The thing is anything beyond that you are spending as much as a car. You are basing your arguments on hypothetical specs for a chip that has not a single feature locked in. No one including myself knows yet what this would look like.
Arguing that it is pointless is ridiculous since 6 companies each spending 1 mill plus and hoarding a chip for themselves is done for a reason. BECAUSE IT IS WORTH IT. They make so much money they don´t need to sell it to split the costs.
The argument of not being able to recover the costs is missing the point. They are buying several chips right now which cost enough to cover development of those chips, company profits for those chips, marketing costs for those chips. These costs are still there. When McDonalds sells you a Happy Meal that Toy is not free. YOu pay for the toy in the price of the happy meal. Same with buying 1 chip instead of 5-6-7-8. Buying 1 chip is cheaper because of efficiencies or NO COMPANY WOULD EVER make their own chip. It would never be worth it to combine chips into one custom chip.
Yes Intel paid less than 3 Billion compensation for 20 years of pushing them out of the market. They didn´t just pay Dell they paid ALL OEMs Toshiba, Lenovo, Samsung, HP, All of them took bribes. These companies sell/sold around 80% of the worlds PCs. AMD was forced to settle in 2007 or 08 when the recession started and they would have gone bankrupt if they hadn´t. Fact is AMD was almost locked out of the market from 1984-2012 while producing a very expensive product and that is where the money problems came from and not because the profit margins were not there. Someone did the math recently and their entire Server chip production comes from a few like 2k wafers a year. Now look at how much profit the server market brings them. That profit is coming from around 100k CPUs.
Quick google
Search: "AMD 7nm wafer cost"
AMD cost per wafer, yields, and cost per good die for 7nm, 12nm, 14nm
https://adoredtv.com/zen2-chiplet-quality-examined-by-cost/
Here are similar prices for Apples wafer and die costs.
https://wccftech.com/apple-5nm-3nm-cost-transistors/
Search: "300mm wafer cost"
Wafer Profits for Pure Play $7.41 per sq inch at .5 micron on a 200mm wafer $53.86 per sq in at <20nm on 300mm. So same square inch produces 8x the profit because of higher costs.
https://www.elektroniknet.de/markt-technik/halbleiter/wafer-preise-unter-druck-158732.html
tautech:
--- Quote from: excitedbox on January 11, 2020, 12:07:26 am ---
Someone mentioned a ~$1300 scope with 350MHz. Hantek has a $600-700 scope with 300MHz. The thing is anything beyond that you are spending as much as a car.
--- End quote ---
Nope, you missed the point.
So does Siglent in the 2ch SDS2000X-E series, for $ 620 you can have the 200 MHz version that can be hacked to 350 MHz....done it so proven.
SDS5000X 4 channel models of 350 MHz are the same HW as 1 GHz @ 5 GSa/s and using two 5 GSa/s ADC's.
The point being SDS5034X @ ~$3.5k with current promo of $2k of free options is effectively just $1500 if you wanted all of them.
Current marketing and technology is giving us deals like never seen before which leaves little fat on the bone or desire to push to higher levels of performance such is the small amount of sales in that sector.
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version