MIPI UniPro is a recent addition to mobile chip-to-chip interconnect technology. It’s got many useful features to meet the requirements of mobile applications. That’s perhaps why Google’s Project Ara has selected MIPI UniPro and MIPI M-PHY as its backbone interconnects.
In this blog post, we describe three differentiating features, benefits and their verification challenges. All the discussion is referenced to MIPI UniPro 1.6.
MIPI UniPro provides six power modes to meet different needs. In SLOW mode, it supports seven gears with operational speed ranging from 3Mbps to 576Mbps per lane. In FAST mode, it supports three gears with operational speed ranging from 1.5Gbps to 6Gbps per lane. Both SLOW and FAST can be coupled with automatic M-PHY’s BURST closure during traffic gaps called AUTO. In the complete absence of traffic hibernate mode is used. All unconnected lanes shall be put in OFF mode. UniPro allows independent power mode settings for both transmit and receive direction.
UniPro allows the dynamic selection of number of lanes, gear and power mode per direction using Power mode change request (DME_POWERMODE) and hibernate state transitions through (DME_HIBERNATE_ENTER and DME_HIBERNATE_EXIT) primitives. MIPI UniPro L1.5 Layer accomplishes these requests through the PHY Adapter Configuration Protocol (PACP) frames of the type PACP_Pwr_Req and PACP_PWR_Cnf. Traffic is paused briefly during the power mode change procedure. Power mode settings are applied simultaneously after completion of power mode change procedure on both ends and traffic is resumed.
This feature allows MIPI UniPro in achieving optimal “performance per watt” of power through setting of appropriate power mode. Based on the application’s data traffic bandwidth and latency requirements, it can scale the number of lanes and operational speed of lanes in each direction dynamically.
Following parameters gives rise to large state space
Functional verification will have to cover the unique combination of all of the above power mode state space (mode x lane x gear). Additionally two more important transition combinations have to be covered:
This would require a constrained random stimulus support. The constrained random stimulus generation is not quite straight forward. It will have to take in to consideration:
Based on above parameters the legal power mode changes will have to be initiated from both the VIP and DUT side.
UniPro allows using multiple lanes (max up to 4) to scale the bandwidth. UniPro Phy adapter layer takes care of distribution and merging of data. During the L1.5 layer’s multi-phase initialization sequence the total number of lanes connected and their physical to logical lane mapping gets determined.
Training sequence identifying the logical and physical lane mapping. Source: MIPI
This feature provides the flexibility in the UniPro’s chip-to-chip lanes routing. Considering the small footprint requirement for mobile hardware this will surely ease printed circuit board designer’s life.
From verification point of view need to cover the following:
Typically through configuration, the number of lanes connected and for connected lanes the logical to physical mapping used needs be randomized. Based on this configuration, the VIP will drive the specified number of lanes and advertise appropriately to the DUT.
MIPI UniPro supports two traffic classes traffic class 0 (TC0) and traffic class 1 (TC1). Traffic class 0 support is mandatory while traffic class 1 support is optional. Priority based arbitration between traffic classes is supported. The MIPI UniPro stack, right from its transport Layer L4 to the data link layer L2, is traffic class aware to provide enhanced Quality of Service (QoS).
At the transport layer level, the logical data connection is the connection oriented port (CPort). It is mapped to either TC0 or TC1. Cports mapped to higher priority traffic class will have precedence over CPorts mapped to lower-priority traffic class. Within a traffic class, segment level round robin is the default arbitration scheme.
To reduce delays and to improve Quality of service (QoS) at the data link layer level, it can insert high priority frames within a lower priority data frame under transmission. This feature is called pre-emption. It’s an optional feature. This concept is extended to other control frames as well for improving the latency and reducing the bandwidth wastage during retransmission.
Composition with pre-emption (Traffic class Y > X). Source: MIPI
CPort arbitration and pre-emption provide fine control over latency of communication. This enables improved QoS. This feature can be used for latency sensitive traffic.
From the verification point of view, we need to address:
QoS feature intent can be verified by measuring the latency on both transmit and receive path of the DUT. This can be done as additional functionality of the scoreboard. The scoreboard can record the time stamp of the messages entering and exiting ports of DUT on both CPort and serial line. The latency of transmit and receive path of DUT can be checked against the desired configured value. Any violations can be flagged as warnings or errors based on the percentage violations.
To ensure that the pre-emption feature is functional, both the legal and illegal pre-emption cases needs to be exercised. Based on the supported priorities table for DL arbitration scheme, there are 18 illegal and 35 legal pre-emption scenarios possible. Both legal and illegal cases must be covered on both transmit and receive path of the DUT including multi-level pre-emptions.
For all these features verification, a well-architected Verification IP plays a critical role. Verification IP with right level of flexibility and control can significantly accelerate the closure of verification.
Authored by Anand Shirahatti, Divyang Mali, Naveen G