Catching up on some of the news from the last twenty days or so, let’s start with Intel: the company’s shareholders have reportedly rejected compensation packages for top executives, including a payout of as much as $178.6 million to CEO Pat Gelsinger.

Advances to Real Intent’s Meridian DFT

Real Intent has announced advances to its “Meridian DFT” multimode DFT static sign-off tool with root cause analysis. Meridian DFT now presents the results as tables, with detailed coverage reporting for violating registers, design instances with the most uncovered faults, and selected test points per test mode. It also annotates coverage debug attributes per test mode for controllability, observability, and stuck-at-0/1 faults. Additionally, the new version includes tables with hyperlinks for cross-probing of nets and instances to the schematic or source code viewer. Real Intent has also made additions to Meridian DFT’s rules, including specialized rules for sequential capture through loops and deep sequences of flip-flops without scan collars, controllability and observability through memories, and advanced connectivity checks.

(more…)

The Linley Spring Processor Conference 2022 – which took place last April 20^th and 21^st – saw the participation of numerous sponsor companies, many of them offering deep learning acceleration solutions. This week EDACafe takes a quick look at the conference content, mostly focusing on some technology trends and some new announcements. Full proceedings of the event can be accessed from www.linleygroup.com, the website of the technology analysis firm now owned by Canadian reverse engineering company TechInsights.

Ever-growing NLP models

In his keynote speech, TechInsights’ principal analyst Linley Gwennap pointed out that language-processing models keep growing at an impressive pace: Alibaba’s M6 has 10 trillion parameters. Model size is limited by training time (compute cycles): for example, training the GPT-3 model using one thousand A100 GPUs takes more than one month. Rapid growth has been achieved by moving to large and very expensive clusters. Recent progress focuses on adding parameters using fewer GPU cycles: for example, Alibaba reports training M6 required only 15% the time of smaller GPT-3. Training can be accelerated through ‘model sharding’, which divides a model across many chips. This requires complex software, possibly with manual assistance. Scaling massive models across servers and racks, sharding requires high-bandwidth connections.

(more…)

Politicians around the world are getting increasingly involved in initiatives aimed at boosting or protecting their countries’ competitiveness in the semiconductor market. This is why ‘geopolitical’ issues make up a significant part of this week’s news round-up. Other updates concern the controversial Google paper on deep learning-based chip block placing, high-NA EUV advancements, and more.

Geopolitical issues: Arm IPO, Indian investments, Alphawave-OpenFive deal

UK prime minister Boris Johnson has reportedly joined a final push to convince Arm – which is currently preparing its IPO – to list on the London stock exchange, as the UK government is concerned over the damage if Britain’s best-known tech company chooses New York for its initial public offering. SoftBank chief executive Masayoshi Son, however, has reportedly described New York’s Nasdaq exchange as “the most suitable” as it is “at the center of global high-tech”.

(more…)