SAN FRANCISCO--Samsung touted its new 32-nm high-k metal gate (HKMG) process at the International Solid-State Circuits Conference here Tuesday (Feb. 21), showing features it hopes might win customers over from a supply constrained TSMC.
While ostensibly presenting its upcoming 32-nm quad-core mobile processor, Samsung spent much of the time presenting its new process, which as well as boasting HKMG, also uses dynamic thermal management and body bias techniques, for a significant 40 percent performance increase over its 45-nm Exynos chip.
Intel and Globalfoundries have been shipping chips with HKMG for a while now, but with Samsung and TSMC joining the fray, the fab playing field is somewhat equalized.
With the industry bracing itself for a tight supply of 28-nm from TSMC, Samsung’s 32-nm process and excess capacity manufacturing facilities could prove an attractive alternative to the likes of Nvidia, or even AMD.
Analysts believe TSMC would give priority to customers like Apple or Qualcomm for the 28-nm process, which would make it difficult for smaller players to get the supply they needed. Samsung is not supply constrained, and the difference between a 28-nm process and a 32-nm process using HKMG is fairly marginal in terms of the end result.
On the other hand, some in the industry still have qualms over using Samsung as a foundry, largely because the firm competes with almost every one of its potential customers.
Samsung's chip-building capabilities were being put under the microscope on Tuesday, with the firm also talking up its dynamic thermal management system for mobile SoCs, as well as power gate technology for every CPU and even inside the GPU, allowing power to be switched off, cutting leakage at the source.
Body Biasing, too, would give Samsung a differentiating factor from the likes of Intel, which doesn’t use the technology. Body Biasing can be used to either speed up a transistor when full throttle is needed, increasing leakiness slightly, or it can slow the transistor down, stemming leakage for a more optimal low power state.
Whether customers will be enticed, however, remains to be seen.
The landscape in advanced node is changing very fast. Samsung will not be a classical foundry (100s of customers) - they will focus on few mega-customers and their HV products.
A long-term objective - already for decades - has been to become larger than Intel. This time they have a shot to achieving that.
Samsung has 40 % of the DRAM market and unlike the smaller competitors actually makes a solid profit in DRAM by staying ahead in technology & yield. It then recycles the learning and resources into processors.
TSMC also has a 50 % market share in the Foundry business.
It remains to be seen who will be the first to fall victim to the complacency & arrogance typical of many market leaders.
Some might even say that one of these two is already showing it.
This is needed. TSMS monopoly is getting over par. We need choices. Yet, who will like to fab with Samsung. You do today and next week they "push your circuit" to the market. It is risky to feed your executor.
I guess the only thing that could hamper Samsung's progress in the foundry business is the fact they compete with almost every one of their potential customers. I do not know how they will square this.
Very interesting as we just posted two weeks ego
a blog by Deepak: Why Samsung will give Morris Chang sleepless nights
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.