The first day of the SOI Consortium’s recent China event – the 7th Shanghai FD-SOI Forum – was full to bursting in every way: the room, the networking, the level of expertise, the in-depth presentations and the overall energy. We covered the Samsung and GlobalFoundry keynotes in our previous post (if you missed it, read it here).
This post will recap the rest of the presentations given during the day. (If your company or institution is a member of the SOI Consortium, you’ll be able to access the full presentations online.)
When it comes to deep insights on China + tech + analytics, and especially with a thorough understanding of FD-SOI markets, Handel Jones is arguably the world’s leading expert. Here are some of the observations he shared. Though the chip industry will see declines across the board in 2019 (he sees 13.5%), he sees a return to growth in 2020. By 2030, he sees it as a trillion dollar market, of which China will have half. AI is a key driver – and will become more prevalent at the edge. Major drivers will include preventative medicine, gaming, NB-IoT and 5G. At the chip level, FD-SOI has a lower cost/chip compared to bulk – you’ve got small chips and high yields. Sensors – especially image sensors – are a key area, and this is another place where FD-SOI is better than bulk. He sees chip shortages in the 2022-24 time frame (as opposed to the current oversupply), so now is the time that China should be establishing large FD-SOI capacity.
In terms of power consumption, computing is easy but data transmission is hard, Ron Martino reminded the audience at the onset. That’s why you need the edge. This is where FD-SOI comes in, and if you want to have leadership, you should be leveraging body biasing, he said. In terms of machine learning, a lot can happen at the edge on the smallest devices. NXP is now shipping a very wide range of products based on FD-SOI, including the i.Mx7 and 8 families and the new RT crossovers. The latest announcement is the i.Mx RT 1100 MCUs, a very low-cost processor solution for high volumes. The i.MX7 ULP is in mass production for wearables, with record low leakage and high performance. The i.Mx8 and 8x are going into a broad range of applications – from retail solutions for automated checkout to pasta makers, and automotive applications for full cockpits with vision detection, as well as things like parking, V2X and in-vehicle monitoring.
Chips built on FD-SOI with eMRAM are in production, said Kenichi Nakano. In GNSS/GPS, Sony is the #1 in lowest power consumption worldwide, thanks to FD-SOI, he continued. They’ve had 70 remarkable design wins, giving them over 50% market share in the sports and health watch markets, he said with a tip of the hat to the FD-SOI ecosystem and SOI Consortium. In GNSS, performance is very important – and now they can do it in water, which is huge. Development cycles are shorter than ever – for the latest chip it started in February 2018 and was in production by the spring of 2019, achieving decreases of 20% in power, 30% in area and 10% in cost. Integrating eMRAM was easy in terms of the design flow and manufacturing, with production yield of 97-100%. So with the GXD5605GF they’ve got the first GNSS chip with FD-SOI/eMRAM/RF in the world and it’s on 28FDS/eMRAM technology. It’s very reliable and very good, he concluded.
At the beginning of this year Rockchip announced the launch of their RK1808, a low-power AIoT solution with built-in high performance (3TOPS) NPU fabbed in GlobalFoundries 22FDX, said Feng Chen. Their clients were very happy that Rockchip delivered the real power and performance numbers they’d promised. Because of the power/performance it delivers, FD-SOI (both 22 and 28nm) is very well suited for AIoT chips, he said. It’s very cost-effective in terms of NRE and die, and there’s room for further savings. While the ecosystem needs a unified push, FD-SOI is good for the market in China, and China has the volumes FD-SOI needs. Rockchip sees particular potential in retail and smartphones.
VeriSilicon CEO Wayne Dai moderated the first panel, asking first why China should adopt FD-SOI. Soitec CEO Paul Boudre said because it is a big, dynamic market (noting that Sony’s first FD-SOI GPS win was in China). Handel Jones said that at the wafer level, there was cost parity, but with FD-SOI chips are smaller and higher yield. The main reason it’s taken so long to get going was IP, but that’s changing now, he added. Dai’s next question was about the top application fields the panelists predicted for 2020. Sony’s Kenichi Nakano said wearables with connectivity, low power consumption, small size and high levels of integration; Rockchip’s Feng Chen agreed. NXP’s Ron Martino said FD-SOI for automotive, machine learning and edge computing was shipping now, with wearables ramping.
The “value” of IoT data is not yet being generated, noted Yi Zeng but AI can help here. The IoT industry needs innovation for both chips and networking. SiPaaS – which stands for Silicon Platform as a Service – as offered by VeriSilicon can help lower the barrier to entry. [In the SiPaaS model, VeriSilicon has its own IP-based core. Based on the company’s advanced chip design capabilities and mass production service experience, it has created a variety of silicon-proven chip design platforms that can significantly reduce the customer’s chip design cycle.] They have FD-SOI IP for NB-IoT, BLE, GNSS and sub-1 GHz. The BLE (Bluetooth) RF IP is a complete offering optimized for low power on GlobalFoundry’s 22FDX. The NB-IoT IP is also optimized on their 22FDX ZSPNano, an energy efficient general purpose MCU+DSP core on 22FDX. And they’ll have results of test chips for GNSS RF IP on 22FDX by the end of this year.
While AI enables products and services, it’s important to plan for security early in the design cycle, said Hassan Triqui. Software is not enough to protect edge-to-cloud. Secure-IC’s hardware security module, Securyzr, is an IP block that can be embedded into every device to answer security functionalities such as root-of-trust and key management. In sleep-mode/tunable cryptography, FD-SOI allows the creation of physically secure systems. (Note that designers are leveraging FD-SOI’s unique body biasing for ultra-low-power deep-sleep modes.) Because safety and privacy require a combined solution, Securyzer is particularly well-suited to IoT chips built on FD-SOI, he concluded, so that IoT adds value to AI, and not just the other way around.
There are four key areas to 5G, explained Michael Reiha: coverage, number of antennas, frequency and traffic density. 5G mmW access architectures are currently inefficient in terms of power and performance, but FD-SOI is ready for 5G access as both an analog and hybrid beamformer. For MU-mMIMO (massive MIMO), the RF front-end modules (FEMs) and transceiver will fully exploit FD-SOI. Sensing, calibration and control enabling hybrid beamforming and multiple users is easy in FD-SOI. The adaptive body biasing on the horizon will reduce power of FEM mixed-signal circuitry, and be a disruptive technology.
As a leader in the automotive market, ST has seen that increased data flows in automotive are driving demand for higher performance and bigger memory in automotive MCUs, said Shan-Lin Liu. ST has taken a unique approach to NVM with embedded PCM (phase change memory) on 28nm FD-SOI. This gives them energy-efficient, high-performance cores with larger NVM memories, and it’s already qualified up to auto grade-0. PCM (vs MRAM) is BEOL. It uses two cells, so it’s more reliable and is good at high temperatures, he said. With FD-SOI, they can go up to 165o, and it’s soldering compliant. The preliminary results of the first MCU chip are excellent. It’s now running in a car, replacing the previous generation 40nm eFlash product.
To fully run artificial intelligence on the edge, research powerhouse Leti is working on an unsupervised learning neural network using advanced FD-SOI and a mix of other technologies. These include embedded non-volatile memory (NVM), 3D integration, and new design tools. Sabonnadière said this new approach is expected to exceed the performance levels of current digital deep learning with neural networks that are capable of handling time-domain signals, sound and speech—and may produce a first “killer app” for advanced SOI. AI will require compact and power-efficient circuits for the inference phase, when neural networks infer things based on new data they receive, close to the end user. The combination of FD-SOI, 3D integration, and NVM opens a path towards dedicated circuits with major performance improvement within the limited power budgets of distributed electronics. In Europe, he noted, privacy concerns are driving the move from the cloud to the edge. On the Leti roadmap, they’ve broken through the 10nm limit for FD-SOI, using strain and body biasing to compensate for transistor mismatch. Also of note: since 2016 Leti has had an ongoing collaboration with SITRI, the Shanghai Industrial μTechnology Research Institute, an international innovation center focused on globally accelerating the innovation and commercialization of More than Moore technologies to power IoT.
Dresden Fab 1, Thomas Morgenstern reminded the audience, is the biggest in Europe, where it is part of the Saxony ecosystem. GF is moving advanced mask-making to Dresden, which is the lead site and Center of Competence for FD-SOI. With the “pivot”, GF is providing platforms. Fab 1 is automotive certified for 22FDX (GF’s 22nm FD-SOI technology), with automotive tapeouts in 2019. “Automotive is a journey,” he said, of continuous improvement, and a mindset: it’s a zero defects culture. The ramp to volume production is well underway, with 26 tapeouts of 22FDX products this year – almost double that of last year. He showed high yield data of about a dozen products, adding that since the beginning of the year every tapeout was first-time right with decreased cycle time. The key specifications for 22FDX with eMRAM for Auto Grade-1 have all been demonstrated, and customer feedback has been excellent.
As you can see, it was a packed day for the FD-SOI part of the SOI Consortium’s Shanghai event. In fact the room was still packed at the very end of the day. Several hundred VIPs then headed out for the ever-popular and festive evening riverboat dinner cruise, where the non-stop networking continued.
A big shout-out to our sponsors and supporters: VeriSilicon, Simgui, SIMIT, Soitec, Samsung, IBS Ion Implant, ShinEtsu, GlobalFoundries and NXP.
The next day of the event was devoted to RF-SOI. That will be the subject of our next post.