You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
Hi all – I'm intrigued by the new RP2040 microcontroller developed by Raspberry. It's the first chip that Raspberry has designed themselves – the SoCs they use on their more powerful boards tend to be from Broadcom.
The RP2040 consists of two Arm M0+ cores. I was surprised that it's made with a 40nm process node. Is 28nm not cheaper at this point, in terms of per chip cost? If you were starting from scratch with a project like this I assumed the smaller nodes were cheaper. I know the wafer cost is supposed to be higher, but I thought the area gains made the chip cost lower.
And since this is an ultra low power part, less than 1 mW at idle, wouldn't 28nm be good for another 40-50% in power savings? I've been amazed lately by some of the improvements to 28nm by TSMC and others, with their third or fourth generation offerings, and recently the (still planar) nodes they're calling "22nm" or similar have PPA numbers that seem like they ought to match first-gen (16/14nm) FinFET processes.
Is there a big non-recoverable upfront cost difference between 40nm and 28nm? I thought the smaller nodes would be the default for anything new, where there's no migration from an existing product.
One of the big differences is in the IP you are buying: it is already taped out in both 40nm and 28nm? If it is, usually the price is not that different. But on the Coretex-M0+ webpage, they only list 180nm, 90nm and 40nm: https://developer.arm.com/ip-products/processors/cortex-m/cortex-m0-plus
If it really has not been previously taped out in 28nm, that would make it a very expensive.
One thing I can tell you is that TSMC has ample 28 capacity. If you remember, TSMC went gate-last HKMG while the other foundries went gate-first. TSMC yielded and the others did not which started the TSMC market share domination they have today.
TSMC became the first foundry to provide the world's first 28nm General Purpose process technology in 2011 and has been adding more options ever since. TSMC provides customers with foundry's most comprehensive 28nm process portfolio that enable products that deliver higher performance, save more...
www.tsmc.com
TSMC became the first foundry to provide the world's first 28nm General Purpose process technology in 2011 and has been adding more options ever since. TSMC provides customers with foundry's most comprehensive 28nm process portfolio that enable products that deliver higher performance, save more energy savings, and are more eco-friendly.
TSMC's 28nm process technology features high performance and low power consumption advantages plus seamless integration with its 28nm design ecosystem to enable faster time-to-market. The 28nm process technology supports a wide range of applications, including Central Processing Units (CPUs), graphic processors (GPUs), high-speed networking chips, smart phones, application processors (APs), tablets, home entertainment, consumer electronics, automotive, and the Internet of Things.
TSMC's industry-leading 28nm process technology mainly uses High-k Metal Gate (HKMG) gate-last technology. Compared to the gate-first technology, the gate-last offering provides more advantages, including lower leakage current and better chip performance.
It would've been feeling just ok, if not better on 65nm SOI if standby power was a concern.
I guess 40nm was just because of it being more mainstream, and less risk for a first chip of the company, along with further performance scaling options.
Raspberry Pi is a weird product, very unlike an average industrial MCU which most micros are. Industrial, and consumer applications are likely secondary for it after the hobbyist market.
