Bit noise: 2 nanometer chip, new cpu-lucke and opencl waulty

Bit noise: 2 nanometer chip, new CPU-LUCKE and OPENCL Waulty

Developer teams of IBM Research in Albany (New York) have made the allegedly worldwide first test chip with 2-nanometer technology on a pilot plant. However, 2-nanometer series production was allowed to start early in 2025. Then, ideally, more than 330 million transistors should fit on only one square millimeter silicon surface.

These were more than 50 percent more than chips such as Apple M1 and Qualcomm Snapdragon 888 from the previously advanced manufacturing technology, the 5-nanometer method of TSMC. However, between the 5- and 2-nanometer manufacturing technology, the 3-nanometer technology is still coming, so that the comparison drawn by IBM is a bit crooked.

Bit noise: 2 nanometer chip, new CPU-LUCKE and OPENCL Waulty

In 2025, the production of 2-nanometer chips started with "Nanosheet"-Transistors work. Whose conductive channels (to see above in the image in cross section) are surrounded by the gate electrode surrounding.

IBM has produced with the 2-nanometer technology novel field effect transistors (FETs), so-called Nanosheet Fets. In which the conductive channel is divided into several finer sheets (sheets), which are surrounded by the gate electrode manner. Nanosheet transistors are a variant of the gate-all-around (GAA) FETs, which also develop Intel and TSMC for 3- and 2-nanometer manufacturing. You should then dismiss the previous Finefets whose conductive channel has the form of one or more fins.

Because IBM no longer operates its own series production, cooperation partners such as Samsung and may also offer Intel the Nanosheet technique as a ready-to-use. The cooperation between IBM and Intel is new. With GlobalFoundries, however, IBM has been working together for a long time, but Globalfoundries has so far no plan to produce finer structures than 12 nanometers. At the Dresden site, Globalfoundries produces a minimum of 22 nanometer structures, namely on silicone-on-insulator (SII) wafers; The technique is already there 22FDX.

Micro Ops Lucke

American researchers have revealed a security chute that affects X86 processors from AMD and Intel and potentially Arm chips. Similar to SpectRe, use a side channel: In order to get on supposedly protected data from other running threads, manipulate the micro-surgical cache. This buffer tank sits very close to the arithmetic factors, is filled with a few thousand already decoded micro operations, called micro-ops (μops / uops). As a result, the computing power of current processors significantly increases, which is why protective measures against the new safety chucks could significantly reduce performance. However, the side channel attack, which the research team baptized a bit dramatically "I Lake Dead μops", was quite complicated and has not been included in the database of Common Vulnerabilities and Exposures (CVES). This suggests that it is considered little threatening; Especially he was allowed to be attractive for attacks on cloud servers and do not significantly increase the security risk with typical desktop PCs and notebooks.

OpenCl in the away

The programming interface OpenCl, with which graphics processors can be integrated as computing accelerators, device in a thread. Although the first version already appeared in 2008 and only in the fall of 2020, the standardization committee in the Khronos Group still adopted OpenCl 3.0. Apple has the technology, however, with a security update for MacOS 10.15 (Catalina) shut down. On Macs you should use the Metal API instead or commands from Apple’s "Accelerate" framework. NVIDIA as a dominant seller of computing accelerators has successfully maintained the in-house CUDA (-X) and shows quite little interest in OpenCl. Thus, as OpenCl examples, especially AMD and Intel remain. But Intel forged with OneEApi a separate iron and with SYCL there is another programming model. This API fragmentation was allowed to quench developers who want to bring their software to trot with computational accelerators. This in turn brakes the spread of apps that afford the numerous special functions of modern processors.

More expensive hard drives

At Svenergpasse and the following moon prices of semiconductors you have already used tooth-ridden. At the beginning of May, hard drives with high capacity from about 10 TByte became significantly more expensive. The cause is probably the commercial start of the crypto food Chia, whose "Proof of Space"-Schreufprozess a lot of storage space occupied. The latter is well meant well and should eat less resources compared to the powerful mining about Bitcoin and EtherEmen – but the world really needs more cryptovia?

The bit noise is also available as a podcast.